Introduction

Female sterility is caused by Fallopian tube occlusions in one of three cases. The treatment methods are associated with significant physical, psychological and financial burdens. Our previously described microstent technology opens up new therapeutic possibilities to restore the lumen of the Fallopian tube without surgery. During assessment of loading behavior, the stent design requires design optimization, particularly concerning the crimp diameter, which is presently constrained by self-contact between stent struts.

Materials and Methods

Finite Element Analysis was performed using the software ANSYS Workbench 18 (ANSYS Inc., Canonsburg, USA). A bilinear constitutive material model considering isotropic hardening for polymer poly(L-lactide) was utilized. Microstent radial force was analyzed as a function of the device diameter up to a minimum diameter of 1.0 mm for three different simplified stent models. To evaluate the stent variations, self-contact, maximum stresses (von Mises stress) and volume of stress above 45 MPa as a safety factor of 1.5 of the yield strength were investigated.

Results

All microstent designs were successfully crimped to a diameter of 1.0 mm using Finite Element Analysis. Stent design versions 1 and 2 were limited to crimping diameter considering self-contact. The wave-shaped cell design of version 3 completely prevented self-contact up to a diameter of 1.0 mm.

Conclusion

Within the current study, we successfully implemented a Finite Element Analysis with a material model for a polymeric self-expanding microstent. By using simplified models, computation time was significantly reduced. Stent design version 3 offers the greatest potential for a self-expanding polymeric microstent for minimally invasive treatment of Fallopian tube occlusions. Further Finite Element Analysis for application procedure and in vitro studies will be performed.

As a result of an ageing society, there are more and more people who are affected by extreme mobility restrictions, such as those caused by a stroke. However, younger people with serious illnesses such as amyotrophic lateral sclerosis (ALS) or multiple sclerosis (MS) are also affected. The use of a brain-computer interface to control an assistance robot could help those people to achieve a minimum degree of autonomy and reduce their dependence on caregivers for simple tasks. In this project we use a capacitive and textile brain computer interface to control an assistant robot by visual stimuli. Since the brain signals are in the microvolt range the noise reduction of external disturbances is of main importance. The system is presented and challenges are indicated.

Introduction

Electrical source localisation is important in biomedical engineering. It is about determining positions and charges of electrical sources in a given space, e.g. body region.

Methods

Electrical monopoles and measuring points for electrical potentials were randomly placed in a defined area. The number of measuring points was at least 4*number of monopoles, as the number of parameters to be determined for the monopole is 4 (3 coordinates and electrical charge). In addition, random starting values were also used for calculations. The potentials of the real monopoles at the measuring points were calculated once at the beginning (forward solution, pot_real). The potentials generated by the monopoles were calculated in each iteration with the aid of a gradient descent approach. The sum of the squared differences was used as the loss function, which is to be minimised (loss=∑(pot_real-pot_guess)^2 ). Using the gradient descent method with L1 regularisation, an algorithm was developed that can approximate the initial values to the positions and charges of the real monopoles in an iterative process. The L1 regularisation was applied to the charge only and should figure out the number of monopoles necessary to model the sensed potential distribution, this is achieved by reducing the charge of superfluous monopoles to zero.

Results

The results show that the correct number of monopoles required for the modeling of the sensed electrical potential, generated by monopoles, can be found for various simple charge and potential sensing electrode configurations, when using L1 regularisation for the monopole charges. This was not possible without L1 regularisation.

Conclusion

The results show that L1 regularisation helps to solve the inverse problem, i.e., finding the number and properties of monopoles required to model the electrical potential of an unknown configuration of electrical monopoles. The approach can be extended to estimate properties in a sparse manner of electrical signal sources.

Introduction

Recently, especially during the COVID-19 pandemic, the number of medical air transports with mechanical ventilation has continued to increase. During mechanical ventilation, some forms of therapy (e.g. high-flow) require an increased oxygen level (or flow rate). It is therefore possible that the oxygen level in the cabin of the helicopter will increase significantly. Higher oxygen levels are associated with an increased rate of combustion, lower ignition temperature and can lead to spontaneous ignition. This risk must be examined to ensure flight and operational safety. In this study, the impact of varying oxygen flow rates in combination with air conditioning settings on the oxygen concentration in the cabin of rescue helicopters (Airbus H145/EC135) were investigated.

Methods

The test setup comprised a ventilation circuit consisting of a gas supply, a mechanical ventilator (Hamilton T1) with respiratory mask, oxygen sensors (PreSens OXYBase WR Blue) and a patient mock-up that was integrated into the helicopter cabin. For the series of measurements, the Hamilton T1 was set to PCV+ mode with a constant oxygen concentration of 100 %. The oxygen flow rates (5 l/min, 10 l/min, 15 l/min), four air conditioning settings (summer, winter, recirculated air, off) during main flight phases (ground-run, forward flight) were systematically varied. The oxygen sensors were placed in four locations (measuring interval: one second): near the pilot, the emergency physician, the ventilator outlet/patient interface, and the rear end of the cabin. Mean oxygen-saturation ± standard deviation was calculated for the evaluation.

Results

An increasing oxygen flow rate was associated with a slight increase in concentration (EC135 vs. H145; 5 l/min: (20.81±0.58) % vs. (20.53±0.62) %, 10 l/min: (20.86±0.60) % vs. (20.71±0.80) %, 15 l/min: (20.94±0.64) % vs. (20.97±1.31) %). In terms of sensor location, we had the highest readings at the patient position (EC135 vs. H145; (21.08±0.79) % vs. (21.64±1.36) %). Furthermore, the worst setting for the air condition was “off” (EC135 vs. H145: (21.36±0.77) % vs. (21.00±1.17) %), while “ground-run” being the worst flight phase (EC135 vs. H145: (21.24±0.47) % vs. (20.95±0.76) %).

Conclusion

Conclusively, these results show only slight increases in oxygen concentration within the investigated ventilation configurations and setup variants. Regarding high-flow therapy in medical air transport, further research is needed to ensure patient safety.

In recent years, Electroencephalogram (EEG) based user authentication systems have gained significant interest as an innovative approach for identity verification. EEGs are considered to be a novel biometric attribute due to the individuality of each person’s cerebral activity patterns. This work explores the feasibility and efficiency of utilizing EEG signals, generated in response to emotional stimuli, for user authentication applications, by implementing Machine Learning (ML) and Deep Learning (DL) approaches. Support Vector Machine (SVM), Random Forest (RF) classifier and 1D Convolution Neural Network (CNN) were employed to evaluate and compare the performance of EEG-based user authentication for two publicly available EEG datasets, namely DEAP and DENS database. The performance of EEG-based user authentication was significantly high in LAHV emotional state for DENS dataset, achieving an accuracy of 99.2 % and 92.59 % with SVM and modified 1D CNN, respectively.

This paper presents a comparison of the blood pressure estimation based on BCG measurements, acquired at two different body locations. The first BCG signal being the proximally measured BCG at the carotid artery, the second being the distally measured BCG at the foot. Parallel to the BCG measurement, the reference blood pressure is measured. Visual inspection of both BCG signals shows a decreased signal quality for the distal measurement. Moreover, in order to estimate the blood pressure based on BCG derived features for beat-to-beat intervals two neural networks are deployed. It is shown that mean MAE is twice as large for the distal blood pressure estimation as for the proximal estimation and that the estimated pressure values of the proximal model show less variation when comparing the reference pressure and estimated pressure visually.

Introduction

The electrocardiogram (ECG) measures the electrical activity of the heart, usually with interference from noise such as baseline wandering or movement artifacts. Therefore, it is important to denoise ECG signals for better clinical insights. Current DAE models have shown optimistic denoising performance. Due to long input segments and several convolutional layers their implementation is challenging on limited hardware. This work describes the implementation of single-hidden-layer running DAE model for real-time denoising.

Methods

The proposed RunDAE uses L1 weight regularization for the hidden layer, which is similar in size to the input layer, and a linear neuron as the output layer. A ReLU activation had been selected for the hidden layer. The regularization parameter λ was heuristically set to 10-6. Einthoven I recordings of 3 minutes duration each were recorded from eleven subjects using an Olimex ECG Shield in connection with an Ardunio GIGA R1 with a sampling rate of 250 Hz and an ADC resolution of 12 bits. Two minutes of each recording were used for training and the last minute for testing. All the signals were filtered using discrete wavelet transform, and then corrupted with different noises (level: 0.3 of signal’s standard deviation). Finally, signals were segmented by a sliding-window into overlapped segments, each with length N=200 samples (approximately 0.8 s). The trained model was implemented in C++ on Arduino GIGA R1. Denoising performance was evaluated using root mean square error (RMSE) and signal-to-noise ratio improvement (SNRimp).

Results

The program used 450 KB (23% of available flash), variables and parameters used 51 KB (10% of available RAM). Calculating one output sample from a 200 sample long input took 1.7 ms, which is less than the available 4 ms sample period. The achieved denoising performance is: RMSE=7.0x10-3, SNRimp=4.6 dB for Gaussian white noise and for mixed physical noise we got RMSE=6.0x10-3 and SNRimp=5.3 dB.

Conclusion

The results show that the proposed model requires only limited hardware while providing adequate denoising performance.

These days, many jobs like working in an office

include sitting in chairs for a long time which could lead to

work-related disorders such as musculoskeletal issues. Inappropriate

postures can cause muscle fatigue in certain regions,

resulting in pain and discomfort. Analysis of different types of

postures to indicate discomfort could help us choose an optimal

posture. This study evaluates five different sitting postures

in an office chair for comfort and discomfort. Each posture was

held for 18 minutes with a two minute break between postures.

Six participants with an equal number of male and female subjects

were chosen. The sitting posture correlates with the distribution

of the weight on the seat, which can be measured by

pressure sensors. The pressure distribution was obtained using

a custom-built pressure mat and the maximum pressure

were evaluated. The McGill Questionnaire construct was used

to find subjective discomfort at each minute. There was no difference

in results for both sexes. Overall, leaning to one side

was felt more comfortable while sitting with a curved back

caused the highest discomfort.

The increased use of face masks for infection prevention has resulted in difficulties in speech understanding. Previous studies, however, focused on isolated vowel sounds.

Our study measured the effect of surgery masks on the acoustic attributes of vowels and compared the effect of isolated to co-articulated vowels.

The voices of 45 male speakers were recorded whilst pronouncing isolated vowels and vowels within the co-articulation in a sentence. All recordings were repeated with the participants wearing a surgical mask. Acoustic factors of length, intensity, pitch, formants, shimmer, and jitter were extracted from the vowels in all the different conditions.

The differences between masked and unmasked acoustics in the co-articulated vowels were found to be smaller than the isolated vowels' acoustic features.

These preliminary results improve the quantification of the effect of hearing a face mask in the real-life production of speech. They might be used for better tuning of acoustic plans in hearing aids to compensate for the loss of acoustic information due to the mask.

Introduction

To improve the decision-making process during bladder cancer surgery, additional sensor measurements give valueable information about the tissue’s mechanical, optical and electrical properties. Through fusing the different sensor domains, a meaningful assessment of the pathological state of the examined tissue can be made. Promising unimodal results have been obtained in the electrical and mechanical domains, which are intrinsically linked by the tissue's cellular structure. Given the limited availability of human tissue sensor data, a transfer learning strategy is employed. This includes the generation of measurement data crucial for pre-training an neural network (NN) architecture for sensor fusion.

Methods

As the aim for bladder surgery is to predict the percentage of tumor cells underneath the sensors, a regression network architecture is developed. Hence, a suitable pre-training dataset needs a continuous number as label. In order to achieve the domain fusion, the first step is to fuse the two most promising sensors: an electrical impedance sensor operating in the kHz frequency range and Water-Flow Elastography, which quantifies tissue stiffness through applying a predefined fluid pressure. Therefore, a test setup is developed to apply different pulling forces to porcine bladder tissue, enabling real sensor measurements under varying strain conditions.

Results

The aquired data demonstrate a direct correlation between pulling force, tissue stiffness, and bioimpedance. Elevated pulling force results in increased stiffness and bioimpedance and mimics the effect of higher tumor cell percentage on sensor data, serving as a surrogate label for tumor cell percentage estimation. This dataset enables the pre-training of the NN architecture, which is essential due to the high dimensionality of the sensor data.

Conclusion

We have established a method for generating porcine tissue measurements under varying strains. Utilizing this data, we pre-trained a regression NN, which is subsequently fine-tuned to estimate the percentage of cancerous tissue, thereby addressing the data limitation challenge.

The human hearing is one of the most astonishing senses, allowing us to receive and recognize sounds in various conditions, including noisy environments and multiple sound sources. A system which emulates the frequency decomposition in the basilar membrane, transduction of the sound wave and nonlinear amplification as well as the encoding in the hair cells provides an environment to study improvements for nowadays technological systems. Here, a system completely in hardware consisting of an acoustic MEMS sensor and additional circuitry is used to emulate the properties of the human hearing such as adaptation, feature extraction and encoding of sound.

Distortions in laparoscopic videos affect surgeon visibility and surgical precision, underscoring the need for sustained high video quality during the procedure. This study presents a real-time laparoscopic video quality assessment algorithm, independent of reference content availability. Luminance, local binary pattern, and motion-vector maps are generated frame-wise from videos chosen from a public dataset. Statistical parameters derived from these maps are seen to effectively discern distortion types and severities. These parameters are used to train an evolutionary adaptive neuro-fuzzy inference system end-to-end with subjective score labels. Training and validation loss curves demonstrate the efficient data fitting capability of the model. Performance comparison with other state-of-the-art methods reveals superior results, with high correlation and low root mean square errors. The model is able to accurately replicate the perceptual opinions of medical experts and non-experts, thereby encouraging future research works in this field for stereoscopic, augmented and virtual reality surgical datasets.

Transcatheter aortic valve implantation (TAVI) for aortic valve stenosis can be seen as a disruptive technology that has become the standard treatment for previously inoper-able patients over the past 15 years. A key component of the development process for new transcatheter aortic valve re-placements (TAVR) is the hydrodynamic testing as part of real-time functional testing. Although, fundamental require-ments, necessary testing methods, as well as functional envi-ronments, physiological and pathophysiological stress situa-tions, are defined, resulting testing parameter ranges are only estimated. For this reason, two clinically established TAVR were tested in different parameter settings representing differ-ent characteristic load situations on a patient. In fact, the two TAVR with different dilatation concepts were tested under dif-ferent cardiac outputs (CO) and the resulting effective orifice area (EOA), the closing and the leakage volume was measured and evaluated according to ISO 5840-3:2021. Regarding the two TAVP designs, it can be shown that the high radial force of the Lotus Valve contributes to an improvement in the seal-ing effectiveness of the TAVP. This is because the annulus can be pushed outward by the stent resulting in a better fit between the TAVR and the vessel wall. Additionally, the lamellar skirt of the Lotus Valve also appears to contribute to improved seal-ing. In the presented experiments, the smooth abluminal-mounted pericardial skirt of the Evolut PRO bioprosthesis does not seem to contribute to additional sealing. The devel-oped test procedure aims to contribute to the establishment of new in vitro standards. Future work must involve applying the testing methods to additional clinically established TAVR as well as new developments.

X-ray diffraction (XRD) is commonly used to analyze phase compositions of crystalline samples. Medical applications include the analysis of biotechnological materials and gall- and kidney stones, where composition can inform pathology assessment. XRD analysis methods like Rietveld refinement requires expert knowledge, and multi-phase sample analysis is especially challenging and time consuming. Large-scale medical and biotechnological experiments can therefore be hindered by the need to perform analysis tasks using XRD. Here, we present preliminary results on an automated convolutional neural network (CNN) based method for sample composition analysis using XRD patterns. It can aid experts' analysis using initial estimations, and enable basic judgements for non-experts. Furthermore, we confirm the intuitive notion that analysis performance degrades with sample complexity through systematic investigation using a synthetic dataset.

In vitro test setups for simulated application testing are necessary for physician training as well as for development and benchmarking of various devices in interventional neuroradiology. Within the current study, we implemented a test setup for in vitro simulated application testing. Trackability in the distal catheter area, which was positioned inside a complex 3D vascular model was analyzed for two neurovascular stent systems (Neuroform EZ 4.0x20 mm and Enterprise VRD 4.5x14 mm) combined with different microcatheters (Excelsior XT-27 Flex, PX Slim Delivery Microcatheter and Prowler Select Plus Infusion Catheter). The described test setup allows for a diffentiation of neurovascular stent systems with regard to the trackability force as a major procedure related indicator for device handling. Therefore, the simulated application system presented within the current work represents an indispensable tool for safely and efficiently developing and evaluating new technologies for interventional neuroradiology prior to clinical implementation.

Introduction

Thin-film electrodes on polymer substrates have become standard in flexible neural interfaces. Longevity is a major challenge in translational research to use these approaches in chronic applications in preclinical and clinical studies. Understanding of all potential failure modes is mandatory for risk mitigation and development of concepts that last for a life-time but at least 5 to 10 years.

Methods

We have investigated electrode arrays after chronic implantation to find out potential failure modes. Results have been obtained from animal studies (ferrets, ECoG arrays) in the central nervous system as well as from human im-plantations in the peripheral nervous system. Moreover, digital holographic microscopy allows for in vitro imaging and quantification of the mechanical displacement. It has been complemented by scanning electron microscopy to get insights into potential changes in the grain structure and material composition during stimulation.

Results

We believe that methods of explantation and preparation of chronically implanted probes strongly influence further analysis steps and need to be identified. In addition, our in vitro studies indicate that oscillations in platinum thin-film electrodes can occur during electrical stimulation with progressing adhesion loss over the stimulation period.

Conclusion

Experiments have been performed to quantify oscillation processes, model stress, size and adhesion force as influenc-ing factors and get deeper insights into these processes. This helps us to further enhance our thin-film electrodes for long-term applications.

Electrical Impedance Tomography is used to image the cross-sectional conductivity of an object. It is clinically used for high-frame-rate imaging of lung ventilation. Most current systems use very limited current injection waveforms and patterns. We have developed a flexible system for researching alternatives. This paper covers our current supply module's design and characterisation. We tested the input filtering, bandwidth, and parasitic capacitance. The input filtering functioned similarly to the simulation of the design. However, the bandwidth was limited to approximately 1 MHz instead of the desired 10 MHz due to excessive parasitic capacitance. We proposed points to redesign that should reduce the capacitance by a factor of 10 and correspondingly increase bandwidth 10X to 10 MHz.

Electrical Impedance Tomography is a non-invasive imaging technique that employs electrical currents to reveal the internal conductivity distribution within the body. The study follows a new Radio Frequency (RF) approach, by using a Vector Network Analyzer to provide a multifrequency measurement between 10 kHz and 1 GHz. A plastic and a metallic object were positioned within a saline-filled phantom to test the EIT reconstructed images at different frequencies.

The results show that based on the image reconstruction, the material's conductivity can be distinguished at specific frequencies. However, beyond 10 MHz, the reconstructed images become less reliable until reaching 1 GHz, where the reconstructed image deteriorates significantly.

While Electrical Impedance Tomography (EIT) has potential in new applications, its efficacy is limited by the ill-posed problem of the reconstruction. Further development and refinement of techniques, particularly in addressing issues that occur at higher frequencies, are required to realize EIT's full potential in various fields.

Artificial intelligence is rapidly transforming healthcare, yet in Europe, there are no standardized regulations for AI-based medical devices under the Medical Device Regulation. This gap necessitates independent compliance verification by medical device manufacturers. This paper addresses the regulatory void by proposing a comprehensive training concept. It equips manufacturers with essential artificial intelligence knowledge and outlines current market introduction challenges. Through pre-tests and training sessions, the concept's efficacy will be assessed and refined. In summary, this work offers insights into AI's transformative potential and regulatory hurdles in healthcare, paving the way for improved standards and practices

Introduction

Deeply-seated implants provide the opportunity to monitor patients’ health parameters over an extended period of time. This feature allows implants to assist the patients’ medical treatment in the course of a disease. Yet, this demands an exchange of information between the implant and medical professionals, enabling the making of informed decisions or adjusting the treatment. While most traditional approaches are based on inductive links, they show limitations on penetration depth for safe exposure levels. Ultrasound, in contrast, enables higher penetration depth in human tissue within safety limits, overcoming current distance-related constraints.

Methods

This contribution presents a proof-of-concept for ultrasonic-based communication using binary frequency shift keying (BFSK). For this setup, two ultrasonic transducers are placed 75 mm oppositely apart with a tissue-mimicking phantom in between, and are operated by two TI-MSP430 microcontrollers configured as sender and receiver, respectively. The mark (binary 1) corresponds to 8 pulses with fm = 2 MHz each, the space (binary 0) to 8 pulses with fs = 1 MHz each. The sender first transmits 20 preambles on which the receiver synchronizes before receiving and decoding the actual payload. Additionally, we monitor the energy demand for decoding messages, determine the bit error rate (BER) and evaluate the maximum data rate (Dr).

Results

The minimum energy demand per bit yields 10.5 µJ, increasing to 40.4 µJ by considering further peripherals. The BER was determined as 0.001, the Dr amounts up to 747 bits/s.

Conclusion

The distance of 75 mm exceeds most reported distances with inductive-based links, whereas the low power demand enables its integration in future implants.

Titanium and its alloys play a vital role in dental implantology, providing a stable foundation for replacement teeth and restoring oral function effectively. Titanium alloys, particularly CP-Ti and Ti-6Al-4V, are widely utilized due to their biocompatibility, corrosion resistance and mechanical properties. However, concerns arise regarding the release of cytotoxic elements and stress shielding effects. Addressing these concerns requires advancements in material design and tribological properties. Research focuses on developing low modulus β-type Ti-based alloys with improved wear resistance and biocompatibility, incorporating elements like niobium, tantalum or zirconium. This review explores the tribological implications of these advancements, emphasizing the need for optimized titanium alloys for dental implant applications.

In this study, a measurement setup was developed to determine the frictional properties between tissue-mimicking and implant materials. Nitinol and Polytetrafluoroethylene (PTFE) were selected as implant materials, while Polyvinyl alcohol (PVA) hydrogel and silicone were chosen as tissue-mimicking substrates and compared to porcine aortic tissue. The results showed a considerably lower friction coefficient between nitinol and all substrates compared to PTFE. PVA hydrogel showed a lower friction coefficient than silicone in contact to all implant materials, but still a higher coefficient compared to aortic tissue. Moreover, parameters such as PVA con¬cen-tra¬tion and aging through storage influenced the frictional properties of PVA hydrogel. Finally, feasibility of the novel measurement setup to measure friction under physiological conditions as well as a comparison with previous studies are discussed.

[Introduction] The operating room (OR) is a complex environment. For the optimal support of the surgical team, surgical con-text-aware systems (SCAS) aim to provide the right assistance at the right time. One of these progressive SCAS was developed within the OR-Pad project, which is located near the surgical field and provides preoperative information, depending on the current surgical phase. However, automatic surgical phase recognition is still under research. In contrast, an Intraoperative Checklist relies on manual input to tick off checklist items, guiding the surgical team through complex interventions. To enable context-aware behavior, e.g. for the OR-Pad, the current surgical phase, retrieved from the checklist, could be provided in the OR network. [Methods] The IEEE 11073 Service-oriented Device Connectivity (SDC) standards facilitate interoperability between medical devices. The Interoperative Checklist and OR-Pad systems were extended with an SDC Service Provider and Consumer component, respectively. Thus, the Intraoperative Checklist provides the current surgical phase as a metric. The OR-Pad subscribes to updates and automatically displays the appropriate information when a corresponding surgical phase is clicked on the checklist. [Results] The extension and interaction of both systems were tested in the Research OR at Reutlingen University for eight simulated interventions, e.g. a robot-assisted esophagectomy. Thereby, both systems require to use the same phase definitions and encodings to ensure the correct interpretation of the provided information. Using standardized nomenclatures, such as SNOMED CT, is indicated to achieve semantic interoperability. [Conclusion] To enable SCAS without the need for sophisticated situation recognition systems, the interoperable provision of procedure-related information from an Intraoperative Checklist is indicated. Thereby, reliable context-aware behavior can be achieved via human situation interpretation and the SDC-based provision of this information in the OR network. The approach can be transferred to other systems to provide optimal support to the surgical team.

The integration of mechatronic systems to improve the control and maneuverability of endoscopes in complex gastroenterology procedures is necessary since manual handling often lacks precision. The result are flexible endoscopy robots.

Therefore, the question arises as to how far these robots have evolved. This basically depends on their degree of autonomy and the typical interventions they are designed to perform.

In this review, we provide an overview of existing flexible endoscopy robotic systems, categorizing their levels of autonomy and intended applications.

Of the 28 systems reviewed, most operate at autonomy levels~0 (no autonomy) and level~1 (robotic assistance), with some progress toward level~2 (task automation) in colonoscopy. For more complex procedures, efforts are made to establish master-slave systems.

Advances in flexible endoscopic robots are aimed at increasing autonomy, particularly for autonomous navigation within hollow organs. The integration of advanced sensor technologies is critical to achieving precise control and steering in future developments, and thus higher levels of autonomy for these systems.

Transcutaneous spinal cord stimulation (tSCS) offers non-invasive relief for chronic pain and improves motor function in spinal cord injury (SCI) patients. However, its mechanisms are not fully understood, and patient-specific factors complicate treatment. aims to understand tSCS better by developing a novel Finite Element Model (FEM) of the human body using CT scans. High-resolution CT scans of three subjects' abdominal cavities were used to create 3D models in Materialize Mimics. After pre-processing in Autodesk Meshmixer, the models were converted into solid CAD objects in Ansys SpaceClaim and combined into a single abdominal model. Ansys Maxwell was then used for simulation, yielding results consistent with known current distribution patterns. Despite promising initial results, challenges remain, including long computation times and mesh errors. Validation is necessary before clinical application. Simulation showed varying current densities across different electrode configurations, with T10 to S2 having the highest and L4 to S2 the lowest. This method, employing Ansys Maxwell, allows easy electrode configuration adjustments but requires optimization to reduce computation times and improve mesh quality. Future work will focus on these aspects to enhance the method's utility and reliability.

Proposal of a Method to Identify Early Onset of Sepsis using

Cardiovascular Vital signs

*Sharon Mgute, Department of Applied Mechanics and Biomedical Engineering, Indian Institute of Technology Madras,

Chennai – 600036, India, Email: sharonmgute@gmail.com (*Corresponding Author)

Sreelakshmi Shaji, Department of Applied Mechanics and Biomedical Engineering, Indian Institute of Technology

Madras, Chennai – 600036, India

Punitha Namadurai, Department of Biomedical Engineering, Sri Sivasubramaniya Nadar College of Engineering,

Kalavakkam – 603110, India

Ramakrishnan S, Department of Applied Mechanics and Biomedical Engineering, Indian Institute of Technology Madras,

Chennai – 600036, India

Introduction

Sepsis is a life-threatening medical condition, characterized by organ dysfunction, resulting from an uncontrolled immune

response to an infection. As it is the leading cause of mortality in intensive care units, rapid and early identification of

sepsis is crucial. Vital signs are strong indicators of sepsis which are obtained non-invasively with minimal human inter-

vention. Although studies have employed vital signs in sepsis detection, the utility of cardiovascular parameters is rela-

tively unexplored in the current literature.

Methods

In this study, an attempt is carried out to identify sepsis using cardiovascular vital signs and machine learning model. The

ability of these parameters in classifying sepsis at two different time points are compared and their importance scores are

evaluated. The parameters namely heart rate, oxygen saturation, systolic, diastolic and mean arterial blood pressures of

295 sepis and 305 non-sepsis patients are obtained from Physionet Computing in Cardiology 2019 challenge dataset.

These vital signs are considered from the 7th and the 9th hour as these time points are reported to be within the early sepsis

stage. Further, these parameters are fed into a lasso regression model to identify sepsis and parameter importance scores.

The dataset is randomly split at a ratio of 70:30 for training and testing, and model performance is evaluated using standard

metrics.

Results

Results demonstrate that the mean value of heart rate and blood pressure measurements are higher in the sepsis patients

compared to non-sepsis at the considered time points. Also, the oxygen saturation level decreases in sepsis. The adopted

model acheives an accuracy of 53% and 59% in identifying sepsis condition using cardiovascular vital signs obtained at

the 7th and 9th hour respectively. Among the cardiovascular vital signs, it is found that the oxygen saturation level has the

highest importance in classifying sepsis and non-sepsis patients in both the considered time points.

Conclusion

This study demonstrates that cardiovascular vital signs are able to identify early sepsis condition. The adopted model

identifies oxygen saturation level to be of the highest importance, indicating their clinical significance in early sepsis

prediction. However, extensive analysis using larger data samples is required.

In the world of dental implants, the essential determinant of success lies in the phenomenon of osseointegration, wherein titanium implants exhibit the ability for robust integration with the surrounding bone and tissue. Beyond the inherent qualities of titanium which make it an attractive candidate as an implant material, attention has shifted towards its’ tailored surface modifications. These methods help the implant manufacturers to modify the implant surface for successful osseointegration and higher chances of enduring implantation. This review explores different surface modification methods and their impact on dental implant success.

Introduction

Fluorescence lifetime imaging ophthalmoscopy (FLIO) allows in vivo measurement of autofluorescence intensity decays of endogenous fluorophores. So far, only devices from Heidelberg Engineering (gold standard) have been used in FLIO research. The aim of this study was to compare the new FLIO system (based on RETImap, Roland Consult) with the gold standard using cuvette measurements.

Methods

We performed FLIO measurements using four fluorescent dyes filled in cuvettes: A (25μM eosin Y solution mixed with 5M potassium iodide solution), B (20μM erythrosine B with water), C (15μM eosin Y with 0.5M potassium iodide), and D (3.3μM eosin Y with water). For the two devices and all measurements, the laser power was 101µW. With the liquid column in focus, alternating between the devices, 20 measurements (approx. 1–3min each) were performed for each dye. Monoexponential fluorescence lifetime approximation was performed using the FLIMX software (www.flimx.de). A rectangular ROI (1000px) was defined for each. The mean fluorescence lifetimes of the ROIs were compared for the four dyes in the two spectral channels, short (498–560nm, SSC) and long (560–720nm, LSC). Statistical analysis was performed using Shapiro-Wilk, Mann-Whitney U, F, and t-tests. Bonferroni correction was used to correct for multiple comparisons.

Results

We obtained for the gold standard (mean fluorescence lifetime ± standard deviation) in the SSC: A|13.1±0.7ps, B|75.4±0.9ps, C|393.4±2.3ps, and D|1282.1±7.0ps. In the LSC: A|42.3±1.0ps, B|86.5±0.9ps, C|397.1±2.5ps, and D|1258.0±7.3ps. For the new device in the SSC: A|14.4±1.1ps, B|75.5±1.0ps, C|381.5±2.3ps, and D|1313.5±7.5ps. In LSC we found A|18.3±1.0ps, B|74.1±1.1ps, C|368.5±2.6ps, and D|1237.8±10.2ps. Mann-Whitney and t-tests revealed statistically significant differences in mean lifetimes between the two FLIO devices in SSC for dye D and in LSC for all fluorescent dyes. There were no statistically significant differences in mean lifetimes in SSC for dyes A, B, and C.

Conclusion

Results show that the devices differ significantly in the LSC and slightly in the SSC. Obtaining compatible results in less than half of the cases tested does not allow the conclusion that the results provided by the new device are compatible with those obtained from the gold standard. It is necessary to perform such a comparison on a larger number of measurements.

Unilateral diaphragmatic paralysis in children with univentricular circulation poses significant challenges, including impaired hemodynamics and increased postoperative complications. Restoring diaphragmatic function is crucial to mitigate these risks. This study investigates the feasibility of automated stimulation of the paretic diaphragm half using electromyography (EMG) signals from the non-paretic side following transection of the right phrenic nerve. Our findings demonstrate successful automated stimulation, suggesting the potential development of an EMG-triggered synchronized unilateral diaphragmatic pacemaker. This innovative approach holds promise as a novel therapeutic intervention for managing unilateral diaphragmatic paralysis in pediatric patients with univentricular circulation.

Artificial intelligence (AI) can potentially support time-critical stroke treatment. In a previous study we already demonstrated the feasibility of deep learning based automatic classification for thrombus detection during thrombectomy, a catheter-guided procedure to remove occlusions of cerebral vessels. However, this method has not yet been tested in real time during an intervention. In this work, we present a setup to integrate AI based support in an angiography suite. A classification PC was connected to the angiography by means of a real-time video connection as well as a research interface for control signals. We found that video conversion in real-time does not affect the classification accuracy in comparison to offline classification of DICOM data. Analyzing 50 video streams of retrospective cases, the system could classify digital-subtraction angiography (DSA) sequences in an average time of 13.3 seconds. This processing time can further be reduced to an average of 7.9 seconds with GPU acceleration. Additionally, the system successfully classified two DSA sequences acquired during real-time thrombectomy, identifying the presence of thrombi in less than 5 seconds. So far, the classification result has only been displayed in the control room of the angiography suite to demonstrate its feasibility. In the outlook, however, we also discuss how the result can be displayed directly on the angiography screen as soon as an ethics vote has been passed.

Introduction:

As a first step towards a new method for non-invasive implant monitoring, this work aims to detect periprosthetic bone loss in an in-silico model of a human thigh with a hip-stem implant. To this end, electrical impedance tomography (EIT) is used. EIT is a non-invasive imaging technique that can visualize conductivity distributions within a body and is here applied to quantify the changes in bone conductivity that result from bone loss.

Methods:

Finite element models of a simplified human thigh with a femur and a hip-stem implant are used to simulate time and frequency difference EIT data. EIT voltages are recorded by 16 electrodes placed around the thigh. The simulation includes bone defects of different size and as a regional increase in conductivity. Convolutional neural networks (CNNs) are then used to predict size and position of the defects. The networks are trained on EIT voltages with different levels of additive white Gaussian noise.

Results:

CNNs can predict the defect size and position with little error and few outliers when simulation data without noise or with a high signal-to-noise ratio is used. At signal-to-noise levels below 80 dB, the errors for the predictions increase sharply. Prediction of the defect position requires a higher signal-to-noise ratio than prediction of the defect size. Both, conductivity changes over time as well as frequency dependent properties are successfully used for prediction of bone properties.

Discussion:

In this in-silico model, CNNs can predict the size and position of a periprosthetic bone defect from EIT voltages in the presence of additive noise. This shows that surface voltage readings that contain the information necessary to detect bone loss. Therefore, the application of EIT for non-invasive hip-stem implant monitoring is feasible.

For applications in the cardiovascular field, the production of tailored biomaterials such as those required for insulations or patches is a key technology. In this work we were able to manufacture thermoplastic silicone polycarbonate elastomer scaffolds with a porous external surface. The samples were produced using the dip coating process. Pores were created using the salt leaching technique. The complete removal of the salt was verified by Raman microscopy. The absence of solvents was demonstrated by biocompatibility tests using an endothelial cell line. The potential functionality of the system was thus demonstrated. The next step will focus on promoting the adhesion of suitable cells for specific applications.

Introduction

Oxygenators are a lifesaving technology used for blood oxygenation and decarboxylation in case of acute respiratory failure, chronic lung disease, and during open-heart surgery. The shape of the priming volume through which the blood flows is determined by the housing and the internal end-surfaces of the parts of the fiber-membrane bundle fixed with an adhesive. The traditional potting process results in a shape, which is associated with stagnation zones promoting thrombus formation and redicing device efficiency.

Methods

An adapted potting process is proposed to fabricate blood compartment with beveled end-faces in a classical cylindri-cally shaped housing. Numerical flow simulations using StarCCM+ software and a porous medium model were used to compare hemodynamics in a classic vs. novel beveled shaped oxygenator. Steady-state flow condition with inlet flow rate of 100 ml/s were simulated. However, before oxygenators were simulated, porosity model was elaborated in two separate simulation for the axial and tranverse flow directions.

Results

Since both oxygenator designs have the same size with 100 mm length and 37 mm diameter of the blood compart-ment, flow fields of both designs show similar averaged parameters with a main range of velocity magnitudes between 0 and 0.3 m/s and an averaged velocity of about 0.065 m/s. However, an analysis of regions with low velocities (stag-nation zone) found that in the novel design the volume of these zone is about ten times smaller (0.19 ml vs. 2.21 ml) if compared with a classic design.

Conclusion

Using a numerical study, we have demonstrated that the novel oxygenator design, which can be fabricated by an adapted potting process, results in optimized hemodynamics.

Electrospinning has emerged as a versatile technique for producing nanofibers for diverse applications. The fibers produced are highly regarded for their biocompatibility, exceptional surface-to-volume ratios, porosity, and adjustable composition properties, making them promising scaffolds for tissue engineering. In the literature, annatto-loaded cellulose acetate has already been successfully used for electrospinning. However, N,N-dimethylformamide (DMF) was utilized as a solvent in the experiments, which is considered a carcinogenic and mutagenic substance. The aim of this study is to analyze whether comparable results can be achieved with a low toxicity solvent consisting of acetone and dimethyl sulfoxide (DMSO). In this study, we investigate the electrospinning process of cellulose acetate (CA) and cellulose acetate combined with annatto extract (CA/A). Initially, various polymer concentrations ranging from 12% to 16% (w/v) were characterized by rheological measurements to determine their suitability for electrospinning. The morphology and diameter distribution of the electrospun fibers were analyzed using scanning electron microscopy (SEM). Contact angle measurements were carried out to determine the wettability of the electrospun meshes. The rheological investigations conducted revealed that increasing the polymer content of CA raises viscosity, while the incorporation of annatto decreases it. Uniform fibers are achieved at polymer concentrations of 14% and above. The electrospun nanofibers exhibit uniform morphology and diameters in the nanometer range, indicating the successful fabrication of annatto-loaded CA nanofibers. The contact angle measurements showed hydrophilic behavior on all investigated meshes.

The incorporation of annatto extract in the electrospun fibers holds promise for various applications, including biomedical scaffolds, scaffolds for cultured meat, filtration membranes and food packaging materials. This study provides valuable insights into optimizing electrospinning parameters for the fabrication of functional nanofibers with enhanced properties.

The integration of Artificial Intelligence (AI) into healthcare represents a transformative shift, offering oppor-tunities for enhancing patient care, diagnostic accuracy, process optimization and treatment pathways. This re-search sets out to forge a strategic management decision support framework for leveraging AI within the healthcare sector, aimed at systematically exploring and integrating AI innovations to bolster the patient health outcomes. By creating a comprehensive categorization system, we at-tempt to navigate the complex array of possible AI appli-cations within the field of healthcare, hence enabling the identification, selection, and advancement of AI-driven initiatives. Through a blend of systematic literature review and expert insights, this study maps possible AI applica-tions across dimensions like ‘medical disciplines’, ‘healthcare processes’, ‘AI research areas’, and ‘user groups’. By reflecting the diverse perspectives, this sys-tem transcends mere classification and stands as a corner-stone for identifying, selecting, and developing AI-driven medical use cases to guide strategic implementations of AI within clinical settings. This multidimensional system offers a blueprint for healthcare entities to strategically navigate the AI landscape, enabling them to make in-formed decisions about technology adoption and change management processes, ultimately leading to improved patient care and operational efficiency.

Introduction

A Monte Carlo (MC) light-tissue interaction simulation was conducted for 7 near-infrared (NIR) wavelengths, 1050, 1200, 1300, 1450, 1460, 1550, and 1650 nm, to assess the penetration depth, pathlength, and detected power to de-velop non-invasive sensing methods. The wavelengths selected are commercially available as narrow-band miniature light-emitting diodes in the absorption band of analytes such as glucose, lactate, and creatinine, and are used on a nov-el non-invasive blood analyte concentration monitor. MC simulation was also run for 660 nm, a wavelength that is well documented in the literature, to assess the accuracy of the model. Results from the NIR simulations were also com-pared to experimental testing carried out on porcine tissue.

Methods

The MC simulations were performed using an adapted version of “MCmatlab: an open-source, user-friendly, MATLAB-integrated three-dimensional Monte Carlo light transport solver with heat diffusion and tissue damage”. A 7-layer tissue model was constructed with 'Stratum corneum', 'Epidermis', 'Papillary-dermis',' Upper-dermis', 'Reticular-dermis', 'Hypodermis', and 'Subcutaneous' layers. The absorption coefficient for each layer was defined with the equa-tions based off Jacques "Optical properties of biological tissues: a review" absorption data found in the literature, and data obtained by experimentally by the authors. The emitter-detector spacing was set at 2 mm in reflectance mode. Each simulation was set for 12 hours. MATLAB was used to interpret and compare results.

Results

A MC simulation for the 660 nm wavelength launched 10e9 photon packets. The mean pathlength was 2.22 mm, the mean penetration depth was 0.855 mm. The maximum penetration depth was 5.706 mm. The percentage of photons collected by the detector was 0.0902 %. The maximum normalised fluence rate (NFR) was 129.379 W/m^2/W-incident located in the stratum corneum, and the maximum normalised absorption (NA) was 234.562 W/m^2/W-incident, located in the upper dermis layer. These values are comparitable to results found in the literature, concluding the tissue model is approprite.

Seven NIR MC simulations launching between 9e9-10e9 photon packets. All results are presented in order of wavelength: 1050, 1200, 1300, 1450, 1460, 1550, and 1650 nm. The mean photon pathlength was 2.330, 2.346, 2.432, 2.435, 2.439, 2.477, and 2.548 mm. The mean maximum photon depth was 0.986, 1.012, 1.057, 1.076, 1.080, 1.107, 1.141 mm. The maximum photon depth was 5.682, 4.699, 5.303, 5.348, 5.147, 6.366, and 5.819 mm. These depths are comparitable in to those seen through porcine tissue during experimental testing perfomrmed by the authors. Thus, the penetration depths seen in these simulations are likely to align with real tissues.

The percentage of photons collected by the detector were 0.0544, 0.0475, 0.0440, 0.0392, 0.0389, 0.0370, and 0.0348%. The NFR was 87.8577, 77.0921, 71.4335, 60.3487, 60.0489, 58.9535, 56.1071 W/m^2/W-incident located in the stratum corneum layer for all wavelengths. The NA was 16.3838, 12.9945, 10.1401, 37.0260, 36.5740, 19.4204, 13.5243 W/m^2/W-incident located in the upper dermis.

Conclusion

The 660 nm pathlength and penetration depth was slightly smaller than the NIR wavelengths, and the NFR, NA, and percent of photons collected by the detector was larger as photons at lower wavelengths carry more energy. The pene-tration depth increases with wavelength as suggested by the increase in penetration depth from visible to SWIR. The percentage of photons detected is less than 0.1%, highlighting the importance for highly sensitive photodetectors with these wavelengths. MC simulations were carried out on NIR wavelengths to better understand the light-tissue interac-tion. The pathlength and penetration depths found will be used to develop sensor emitter-detector separation and de-vice selection, and post processing algorithms for a non-invasive blood analyte concentration sensor.

Paralympic cross-country skiing is a competitive and physically demanding sport developed for individuals with physical disabilities. In addition to good training and endurance, sports equipment is a key factor in achieving success. The design of sports equipment must be customized to accommodate specific impairments. Furthermore, biomechanical and neurophysiological factors need to be considered when designing equipment such as ski sledges. Among other neurophysiological factors, muscle activity, typically measured using electromyography (EMG), plays a crucial role. However, due to the high level of dynamic movement in the sport, EMG measurements are not always feasible. This study explores the possibility of estimating EMG data using neural networks and acceleration data. A feedforward neural network model was created and trained to predict upper body muscle activation from acceleration data. Validation of the model using statistical metrics yielded promising results, suggesting its effective use in predicting muscle activity. This research sets the stage for enhancing understanding and optimizing equipment in Paralympic cross-country skiing, ultimately enhancing the performance of para-athletes.

Cast polyurethane films made of an aliphatic polycarbonate-based thermoplastic polyurethane (Carbothane PC) and poly(ether urethane) (Pellethane) were manufactured and their biostability under a static strain rate was investigated in vitro in comparison to unloaded samples. The PU films were stretched to an elongation of 300% and exposed to hydrogen peroxide/cobalt chloride (H2O2/CoCl2) solution for specific periods of time (up to 63 days). Unstrained PU samples were treated in the same way for comparison. The samples were observed for surface degradation via scanning electron microscopy and the bulk erosion by measuring the weight difference. The surface roughness greatly increased in strained Pellethane with scanning electron microscopy (SEM) evidence of deep cracks and holes or ragged and stretched fractures perpendicular to the direction of stress. Carbothane showed a lower degradation compared with Pellethane under mechanical constraint.

The flow control of inspiratory valves for mechanical ventilation is a crucial element of the overall functionality as it mainly governs precision and safety. However, control tuning can be challenging as the widely used magnetic actuation principle comes with a hysteresis shaped characteristic. As a result, the production of ventilators is highly dependent on specific valves and therefore vulnerable if supply chains are interrupted as during the pandemic. An approach which is capable of providing a reliable control for a variety of standard solenoid valves would be beneficial. Therefore, this work examines different models of hysteresis and their precision when being applied to actual measurement data of high pressure servo valves as a foundation for a following model-based control which can be either feedback or feedforward. Regarding the precision, the Prandtl-Ishlinskii (PI) model outperformed alternative approaches (MAE below 2.5 L/min) for the chosen parameters which is why the model is described for fitting both the forward and inverse case.

Introduction

The cardiac magnetic field ranges from 20 to 80 pT, which is several orders of magnitude smaller than the Earth's magnetic field or the magnetic fields found in urban environments. Because of the weak signals, the cardiac magnetic field can only be measured using highly sensitive sensors such as superconducting quantum interference devices (SQUIDs), which have been used since the 1970s. The need to cool the sensors to around -269 °C to achieve the superconducting properties and the additional use of a magnetic shielding chamber have been barriers to the widespread use of SQUID-based magnetocardiograms. Today, other quantum sensors, such as NV centre-based sensors, do not require cooling and a magnetic shielding chamber, but are not widely used due to insufficient knowledge about quantum sensors in the industrial sector.

Methods

In order to improve the understanding of quantum sensors and their potential, functional and macro models are to be created that support the visualisation of their use in the respective field of application. In addition to numerous industrial applications, there are use cases in the medical field, such as magnetocardiography (MCG).

3D transponder coils are simulated to realise the three-dimensional magnetic field of the heart. The transponder coils are arranged in a two-dimensional grid and allow for an individual three-dimensional magnetic field to be generated. The simulation is carried out on the basis of the Biot-Savart law using the BSMag Toolbox in Matlab. The toolbox offers an effective solution for calculating the magnetic flux density in magneto-static approximation, which is generated by a current carrier.

Results

Initial simulations with only the in the z-direction coil active show a good reconstruction of the z-component of the magnetic field in the coil plane. Above this plane at a distance of 50 mm, equivalent to the distance from heart to MCG, the strength of the magnetic field decreases significantly. In addition, the grid-like arrangement of the coils produces a more discrete distribution of the magnetic field compared to the real heart.

Conclusion

The presented model for the simulation of the magnetic field of the heart shows a promising approach but has to be extended. In addition to the three-dimensional excitation of the transponder coils, other arrangements of the coils should be taken into consideration. Furthermore, by switching to a dedicated magnetic field simulation software, additional effects could be included to receive a more realistic simulation.

In laparoscopic liver surgery, image-guided navigation systems provide crucial support to surgeons by supplying information about tumor and vessel positions. For this purpose, these information from a preoperative CT or MRI scan is overlaid onto the laparoscopic video. One option is performing a registration of preoperative 3D data and 3D reconstructed laparoscopic data. A robust registration is challenging due to factors like limited field of view, liver deformations, and 3D reconstruction errors. Since in reality various influencing factors always intertwine, it is crucial to analyze their combined effects. This paper assesses registration accuracy under various synthetically simulated influences: patch size, spatial displacement, Gaussian deformations, holes, and downsampling. The objective is to provide insights into the required quality of the intraoperative 3D surface patches. LiverMatch serves as the feature descriptor, and registration employs the RANSAC algorithm. The results of this paper show that ensuring a large field of view of at least 15-20% of the liver surface is necessary, allowing tolerance for less accurate depth estimation.

One of the major challenges in using magnetic nanoparticles in biomedical applications is still the accurate steering and trapping of these particles at a desired location, like at a tumor. Halbach arrays are well-established here as they generate the strongest magnetic force. However, these arrays are mechanically unstable, and thus, their potential to attract magnetic particles is often investigated fully numerically. Hence, this paper provides instructions on how to easily produce Halbach arrays using permanent magnets and glue. In total, three Halbach arrays of five, seven, and nine magnets were produced and analyzed regarding their potential for capturing magnetic particles in a background flow, numerically using COMSOL Multiphysics and in measurements. Furthermore, the Halbach arrays are compared with two other array configurations with all magnetizations in the same directions and alternating once. Overall, the simulation results were in good agreement with the measurement results. The amount of captured particles was significantly higher for the Halbach arrays with 6.2 % in simulation and 8.2 % in measurements, compared to the array with alternating magnetization direction with 6.0 % and 6.6 % captured particles, respectively. Therefore, using Halbach arrays to capture or steer magnetic nanoparticles in a background flow is strongly recommended.

Introduction

Diabetes mellitus, a chronic condition impacting more than 422 million individuals globally, poses a significant public health concern due to its severe complications.[1] Among these, diabetic foot syndrome (DFS) stands out as a prevalent and serious secondary condition, arising from diminished foot sensation and potentially culminating in irreversible consequences such as amputations. Presently, the treatment approach for DFS focuses on wound immobilization and offloading, yet it faces limitations in assessing crucial medical indicators between appointments and impeding further deterioration.[2]

Methods

To overcome this constraint, a smart wearable monitoring and feedback system has been developed to complement con-ventional DFS care.[3] This system comprises a miniaturized flat PCB sensor component connected via a flexible bridge to a communication unit, seamlessly incorporated into the DFS foot dressing. Continuously measuring tempera-ture, humidity, and pressure in the wound area, the sensor promptly delivers active feedback to the patient's smartwatch in instances of wound overstressing or significant alterations within the wound area. Utilizing flexible and stretchable soft materials is essential to seamlessly integrate the sensor system onto the skin, thereby improving user-friendliness and ensuring precise data collection. To handle these enhancements a pressure sensor has been developed based on thin, flexible, and stretchable substrates.

Results

In a randomized trial with twenty DFS participants, the system was evaluated. When participants surpassed the prede-termined pressure threshold, they received audio-visual alerts on their smartwatches. This biofeedback helped compen-sate for their reduced foot sensation. Consequently, participants adjusted their behaviors, reducing incorrect pressure on their feet and potentially enhancing wound healing.

Conclusion

The case-by-case analysis in this pilot study reveals the suitability of the sensor system for sensor-based DFS treatment but highlights areas for improvement in usability and sensor accuracy. While the current technology functions, integrat-ing flexible sensors could improve overall effectiveness.

References

1. B. Najafi and R. Mishra, Medicina 57, 377 (2021).

2. D. Hochlenert et al. Diabetic Foot Syndrome: From Entity to Therapy. Cham: Springer, 2018

3. D. Hochlenert, C. Bogoclu, K. Cremenns, L. Gierschner, D. Ludmann, T. Tromp, A, Weggen, H. Otten, Sensor-kontrollierte Wundtherapie beim DFS—Das iFoot-Projekt. ALPHA Informations-GmbH, 2022

Introduction

Capsule endoscopes have recently become an important alternative to traditional tethered endoscopes as an inspective and diagnostic method of the entire gastrointestinal tract without causing any discomforts or injuries to the patients. Liquid biopsy provides information about the microbiome that can help in the diagnosis and treatment monitoring of many health issues such as obesity, inflammatory bowel disease (IBD), diabetes, ulceration, and cancer. In this work, a capsule endoscope will be developed to sample a liquid biopsy out of the small intestine, where it is usually difficult to reach through traditional tethered endoscopes.

Methods

A 3D multilayer integrated system with opening and closing elements, channels, a printed microheater, a filter mem-brane, and a vacuum chamber has been built of laser-structured PMMA layers bonded together using pressure-sensitive adhesive. The opening element is made of wax, which melts when the microheater is turned on, allowing the liquid to flow through the filter membrane and to be aspirated into the vacuum chamber. Afterwards, the closing element made of a swellable material (cellulose) swells and blocks the liquid path into the chamber.

Results

The microheater is operated by applying 1 V and 0,18 A for 3-5 seconds, heated at 100 °C from a start temperature of 37° C, which leads to an instant melting of wax and opening of the opening element. Within 5 seconds, the vacuum chamber (100 mbar) aspirates the liquid and a sample of 150 µL is collected. The closing element sucks up the liquid and swells, closing the fluidic path into the sampling chamber within 20 min. The filter membrane of a pore size of 180 µm is integrated to sort out the stool particles.

Conclusion

The sampling mechanism showed full in-vitro functionality. Next steps will be to integrate it in a capsule set-up and test the operation in-vivo.

Introduction

Ocular electrical stimulation (oES) holds great potential for various vision restoration applications. Individualized oES therapy focusing on the retina as the primary stimulation target requires current flow simulation. This work compares forward simulations of oES using finite element (FE) models with different mesh resolutions to find optimal model setups.

Methods

High- and low-density head models are constructed using an MNI template comprising 35.2 million and 13.6 million FEs, respectively. Tetrahedra with average side lengths of 0.4 mm and 0.2 mm represent the retina as the target compartment in the two models. We use an impressed current of 1 mA distributed at eight electrodes (ø 1 cm) positioned around the eye and one occipital return electrode (5×5 cm²). Relative magnitude (MAGrel) and relative distance measure (RDM) compare the current densities (CDs) between the two models in elements on the posterior pole of the retina, as region of interest (ROI). Further, differences in the 95th percentiles of CDs within the ROI are evaluated by their area, center of mass (COM), and orientation of eigenvectors from a principal component analysis (PCA).

Results and Discussion

The average and standard deviation of CD magnitude in the ROI is 0.173 ± 0.049 A/m2 for both models. MAGrel is 0.001 and RDM is 0.037, 0.012, 0.047 in x, y and z directions. The 95th percentiles of the models show great overlap with a COM distance of 0.38mm and an area difference of 2.78 mm2. The three PCA eigenvectors of the percentiles show a small orientational deviation with a cosine similarity > 0.99.

Conclusion

Although the models differ considerably in the number and size of elements, their influence on the resulting, clinically relevant CD distributions in the target ROI is marginal. The lower-density model appears as a reasonable choice for future oES simulations to decrease the model size and computational effort.

Introduction

Polymeric magnetic microspheres (MMS) can be used for multiple applications in medicine and biotechnology, including drug delivery, hyperthermia or immunomagnetic separation. Depending on the application, the MMS need to match specific requirements regarding their size, magnetic properties and antigen binding capacity. Therefore, we are developing size-controlled PLGA and PLA-MS with oleic acid coated magnetic nanoparticles (OA-MNP) and surface-conjugated antibodies.

Methods

Microspheres were produced by an emulsion-evaporation method. Polymer, MNP and drug within an oil phase are homogenized with an aqueous PVA phase. The solvent evaporates out of the droplets and solid MMS are formed. Synthesis parameters were varied to study the tunability of MS size utilizing static light scattering. For incorporating hydrophobic MNP into the MMS, we established an oleic acid coating. Distribution of OA-MNP in MMS was investigated with SEM on focused ion beam cross-sections of MMS. VSM was used tu study the mag-netic properties of MMS. Heating and drug release behaviour of Camptothecin was investigated. Last, antibody conjugation was evaluated using protein A and the biotin-avidin adsorption mechanism.

Results

We found the MS size to depend mainly on homogenization speed and PVA concentration, leading to diameters between 0.5 and 6 µm. Oleic acid coating enables monodisperse suspension of MNP in organic solvents and a homogenous distribution of MNP in the MMS for MNP concentrations up to 33 wt%. MMS can be heated to 43 °C in 100 to 540 sec, depending on MNP content. Drug release showed a burst type kinetic for 37 and 43 °C, whereby Camptothecin was released in its active lactone form, shown by HPLC analysis. Antibodies were im-mobilized on PLA microspheres and confirmed by optical measurements (ELISA).

Conclusion

We developed a toolbox of MMS that can be adapted to several applications by tuning their size, incorporating magnetic nanoparticles and conjugating antibodies to their surface.

Introduction

Cryopreservation plays a crucial role in various fields, including regenerative medicine and cellular therapies. The selec-tion of an appropriate cooling rate in the freezing process is crucial for minimizing cell loss and optimizing therapy efficacy. In commercial cryopreservation, two primary cooling methods are applied: Passive cooling devices regulate the cooling rate by utilizing insulated alcohol baths or thermal insulation material within a -80°C freezer, while active cooling devices usually pump vaporized liquid nitrogen into the sample chamber for gradual cooling. We aim in this work to study different designs of a hybrid method through assuring a cold environment using a mechanical freezer and controlling heat extraction using thermoelectric elements. The designs should offer different cooling protocols and monitoring options for the cooling process.

Methods

The main blocks of the planned prototypes are the power supply and the control unit, which function outside of the cold environment, and the sample chamber consisting among other components of sample holders, a filling medium, insulation and Peltier-elements. Computer simulations are used to analyse and optimize the effects of various design parameters (e.g., geometry, materials of components, heat transfer by Peltier-elements) on temperature fields with respect to efficient heat transfer within the devices and controlled temperature profiles in the stored samples. Different heat sink types are studied for dissipating the heat of the thermoelectric elements. Prototypes are used to verify performed optimization steps and in-silico models.

Results

First simulations show realistic temperature distributions and provide valuable insights for the design of prototypes for effective and efficient temperature control. Simulated/Recorded temperature profiles in and close to the targeted samples serve as useful indicators for optimizing setups and protocols.

Conclusion

This study will be the corner stone in establishing a new generation of cooling devices, offering more flexible and reliable protocols whilst avoiding the high cost of the LN2 infrastructure.

Introduction

The cochlear implant (CI) is a widely used hearing aid in which a microelectrode array is surgically inserted directly into the cochlea where it remains ideally for life. However, statistically, 40% of CI devices fail after implantation and there-fore require re-implantation. A potential complication accompanying CI implantation are inflammation reactions. Therefore, the information about inflammation is crucial for timely admistration of anti-inflammatory medication. Moleculary imprinted polymer (MIP) sensors on a CI electrode are promising candidates for this inflammation detection. For in-vivo applications, the MIPs must meet the requirements of biocompatibility, biodegradability and electrical conductivity. In the presented study, conductive MIPs electrodeposited on a platinum electrode and specific for biotin, used as a surrogate template for interleukin-6, are investigated. The MIPs are first analyzed and subsequently degraded in a controlled manner using impedance measurements at varying electrical potentials.

Methods

PEDOT was selected as the conductive polymer and deposited by cyclic voltammetry. For the biotin sensing, the MIPs were washed in acid-base solution and then impedance measurements were conducted in PBS with and witout biotin. Since PEDOT is not biodegradable, the polymer was electrochemical degraded in PBS by additional impedance measurements, where the applied voltage was varied to modulate degradation. Finally, to investigate the polymer biocom-patibility according to ISO guidlines, the solution containing the degraded monomer molecules was analysed by FTIR.

Results

When impedance measurements were performed, the change in impedance of the MIPs demonstrated a successful incorporation of biotin. With decreasing potential, fewer dissolved polymers and more degraded monomer molecules were detected. Below the potential of 205 mV, only dissolved monomer molecules were obtained, which enables renal clearance. Biocompatibility testing revealed a high biocompatibility for both the polymer and the solution with dissolved monomer molecules.

Conclusion

Based on these findings, we have developed conductive, biocompatible and controllably degradable MIPs capable of detecting biotin.

In this work, an existing experimental setup ("three-bracket-model") for mechanical stress analysis of orthodontic archwires is augmented with an algorithm that automatically finds a neutral position in which the wire transmits as little force and torque as possible. This requires traversing the drives of the setup to an optimal position. The algorithm is first tested with the help of a simulation-based software-in-the-loop framework and then implemented on the real setup. Experimental results show massive time savings compared to manual alignment.

In recent years, the popularity of veganism has surged, prompting a growing interest in its health implications. However, while vegan diets offer numerous benefits, they may also pose challenges due to potential deficiencies in essential nutrients and the presence of anti-nutrients in plant-based foods. This paper addresses the need for a comprehensive understanding of nutrient interactions, particularly those involving anti-nutrients, through the creation of a Knowledge Graph. Utilizing Natural Language Processing techniques, essential nutritional entities were extracted from unstructured texts, and a custom Named Entity Recognition model was implemented. These data were then used to construct a Knowledge Graph using Neo4j, a leading graph database. Despite the potential of the developed Knowledge Graph to visualize nutrient interactions, limitations persist due to the scarcity of available data on anti-nutrient interactions. Nevertheless, the Knowledge Graph provides a promising avenue for exploring and understanding the complexities of nutrient interactions, facilitating future research and informing dietary choices.

Inpatient care facilities globally are facing a critical short-age of staff, posing significant challenges to resident well-being and care quality. This issue is further compounded by demographic shifts and increasing care demands. While technological advancements offer promise in alleviating nursing staff burdens, their effective integration remains complex, with nursing staff acceptance playing a pivotal role. This paper describes a systematic approach designed to streamline the process of identifying, categorizing, and prioritizing suitable technologies in inpatient care set-tings. By taking into account the specific needs and re-quirements of nursing staff, this approach, validated through a comprehensive case study, aims to facilitate targeted technology adoption, thereby contributing to the successful digitization of this occupational domain.

Background:

The diversity of blood glucose meter (BGM) models available in the Polish medical device market hampers self-monitoring blood glucose levels by diabetic patients with commercially available devices. In this cross-sectional study, we analyzed a commercial database of a medical equipment wholesaler regarding the availability and technical specification of different blood glucose meter models.

Methods:

We collected information on commercially available blood glucose meters from a medical equipment wholesaler's commercial database. Blood glucose meters from five manufacturers/suppliers were compared: Accu-Chek (Active, Aviva, Instant, Mobile, Guide, Performa), OneTouch (Ultra Plus Reflect, Select Plus, Select Plus Flex, Verio Flex, Verio Reflect, Verio IQ), Freestyle (Precision Neo, Optium Neo, Libre, Libre Pro, Lite, Freedom Lite), Contour Plus (One, Elite, Link, Contour Next, Contour Next One), and Glucomaxx (BT, Connect, Glucosense, Glucosense Pro, ixell, ixell audio, ixell pro). Technical parameters were noted from user manuals, and comparative statistical analysis was conducted.

Results:

We developed an application to provide knowledge and assist in selecting the most appropriate BGM model for individuals suspected of having been diagnosed with type 1 or type 2 diabetes.

The selected parameters were: blood sample size, measurement time, mass, memory for storing results, measurement range, and average measurement values. All blood glucose meters readings were highly correlated with laboratory measurements; however, venous blood glucose level readings showed slight differences, especially in higher blood glucose levels. Examining the parameters of different models revealed that their key functionalities and features are quite similar. Most devices offer similar measurement accuracy, operating speed, and environmental adaptability.

Conclusions:

Independent comparison of all BGMs and analysis of application functionality were conducted on a group of 30 individuals between 18 and 79 years of age. Tests showed that the application is stable, fast, and easy to use. Users appreciated the intuitive interface and efficient data processing and generating recommendations.

Recent studies show the feasibility of using local remote photoplethysmography (rPPG) for non-contact blood perfusion assessment by creating spatial blood perfusion maps. While global rPPG has been widely studied for its robustness, e.g. for non-contact measurement of heart rate, local analyses pose greater challenges in terms of noise suppression and thus reliability. In this paper, the effect of temperature and illumination changes on signal-to-noise ratio (SNR) perfusion maps is analysed. The results show the importance of consistent temperature and controlled illumination for improving SNR and ensuring reliable blood perfusion measurements using rPPG. This emphasises the need for standardised external conditions for accurate interpretation of the results and medical applicability.

Orthopaedic insoles are necessary for correcting foot deformities and providing customized support. This paper investigates the optimization of insole design through computational methods, with a focus on generative design techniques. By integrating advanced design methods with biomechanical analysis, this study aims to develop customized insoles that effectively treat foot pathologies. To adequately account for all relevant forces acting on the foot during gait, a biomechanical loading model is created. In addition to vertical forces, the model includes horizontal forces (anterior-posterior and medial-lateral) that occur parallel to the walking surface. Generative design can optimize the insole design to distribute pressure and support areas of the foot appropriately by creating functionally graded lattice structures. The study examines the potential of generative design in creating insoles and the impact of the load model on the design outcome. The design process includes dynamic pedography and gait analysis to ensure that the insoles are tailored to the patient's individual needs. Future research challenges include incorporating horizontal forces and minimizing mass while maintaining support. The study highlights the potential of computational methods, such as generative design and artificial intelligence, in optimizing the design of orthopaedic insoles to ultimately improve patient comfort and mobility.

Introduction

Accuracy assessment of electromagnetic source imaging techniques requires physical or simulated head models. A physical head phantom can provide a ground truth verification of the source localization. However, it remains challenging to construct a head phantom with realistic geometric shapes and structures from mechanically stable synthetic materials exhibiting electrical conductivity values similar to the standard values of human head tissues found in the literature. Here, we present the construction of a novel realistically shaped physical three-layer head phantom with embedded electric current dipoles.

Methods

The head phantom is based on the ICBM-NY model, which represents the average head anatomy of adult subjects. The scalp, the skull, and the intracranial volume of this model form the three compartments of the phantom. Casting molds for the three compartments were printed with production-oriented geometric adaptations for additive manufacturing. Electric current dipoles are designed to fit through the neck, be positioned inside the intracranial volume, and are activated e.g. with sinusoidal waveforms from constant current sources.

Results

The scalp layer is made of 2 wt% agar hydrogel with a conductivity value of 0.314 S/m. As for the skull layer, it is cast with gypsum and shows a conductivity value of 0.0017 S/m. To ensure mechanical stability, thin areas of the temporal bone of the skull are thickened outwards to reach 5 mm thickness. The intracranial volume is filled with 0.17% sodium chloride solution with an obtained conductivity of 0.332 S/m. The head phantom is anatomically realistic and includes simplified features such as neck, ears, nose, eyes, and facial bones.

Conclusion

Our results show the applicability of the selected materials and construction techniques for a three-layer head phantom. These materials can mimic human head tissues with the required conductivity values. Our three-layer head phantom is suitable for metrological verification of electromagnetic source imaging techniques.

Synthetic muscles that recapitulate native skeletal muscle contraction and force generation are highly sought after for various applications such as actuators or for regenerative therapeutic interventions, as for instances muscle replacement. We utilize two-photon stereolithography to 3D print synthetic muscles from bovine serum albumin to form a contractile hydrogel structure with tailored mechanical properties. Using a custom microscale tensile strength device, we quantify force-length relations and elastic properties in these 3D printed fibers. We observe that the pH-dependent contraction of these synthetic fibers results in active forces comparable to those occurring in native muscle fibers. Manipulating the length of the synthetic fiber affects force production, following a parabolic force-length relationship similar to native muscle fibers, characterized by an ascending limb, a plateau, and a descending limb as the length increases. Hence, our two-photon crosslinked synthetic fiber can replicate the contractile dynamics of skeletal muscle tissue. This opens new avenues to further explore contractile synthetic materials, as well as their future refinement by integrating myocytes for a more realistic synthetic muscle fiber.

With growing interest in laboratory automation and high-throughput systems, the amount of generated experimental data is rapidly increasing while analysis methods still require many manual work hours from experts.

This is prevalent in X-ray photoelectron spectroscopy (XPS), where quantification is a complex, time-consuming, and error-prone task. We therefore propose a neural network-based workflow to make this process more approachable.

As training data availability ranges from insufficient to non-existent, our workflow is able to create a synthetic dataset containing XPS signals and corresponding area percentages based on binding energies supplied by the user. As a result, no previous measurements are needed. After training on the synthetic data, the neural network can predict area percentages of the known binding energies with high confidence.

This workflow can therefore be adapted for XPS quantification tasks to filter significant data and supervise processes. Moreover, this enables non-experts to analyze spectra and can help experts to reduce focus on important spectra.

Extracting structured information from scientific works is challenging as sought parameters or properties are often scattered across lengthy texts. We introduce a novel iterative approach using Large Language Models (LLMs) to automate this process. Our method first condenses scientific literature, preserving essential information in a dense format, then retrieves predefined attributes. As a biomedical application example, our concept is employed to extract experimental parameters for preparing Metal-Organic Frameworks (MOFs) from scientific work to enable complex and information-rich applications in the biotechnology-oriented life sciences. Our open-source method automates extracting information from verbose texts, converting them into structured and easily navigable data. This considerably improves scientific literature research by utilizing the power of LLMs and paves the way for enhanced and faster information extraction from extensive scientific texts.

Patient-ventilator asynchronies occur during mechanical ventilation when there is a mismatch between the patient's needs and the ventilator's settings. Ineffective Efforts during Expiration (IEE) is such an asynchrony, which if undetected can cause patient stress and prolong the ventilation duration. Current approaches to detect IEEs only target adult patients and are not directly applicable to newborns. This work explores the use of Recurrent Neural Networks (RNN) and Long Short-Term Memory (LSTM) models to detect IEEs in neonates. An attention mechanism is employed to additionally offer a visual explanation for the network's classification. All of the tested attention RNN architectures yield an accuracy of over 90%, with LSTMs performing slightly better than simple RNNs. A user study is conducted to evaluate the usability of the employed explainability method. The results show that the used visualizations are intuitive to understand, however the network's attention itself can be misleading in certain cases.

The development and discovery of new drugs and therapy products in regenerative medicine are carried out through multiple stages which involve several strict quality protocols. The use of cell-based model systems is the most relevant alternative to deviate from animal testing. However, the underlying biomanufacturing process in a 3D environment can be prone to failures or unexpected behaviors such as cell growth anomalies or inconsistencies in cell aggregation, which can seriously affect the quality of the result. Therefore, it is necessary to establish a non-invasive and continuous monitoring mechanism that minimizes contamination risks, allows taking early decisions and enables large-scale production in the long term. The results of our embedded machine learning approach show that it is feasible to train machine learning models that can operate on a resource-limited embedded board with acceptable prediction accuracy for the estimation of the average size of growing stem cell spheroids. Further experiments are needed to investigate the full information potential of the recorded process data.

Introduction

In CT imaging higher radiation doses are needed. If doses are lowered, the image will become noisier. Therefore, techniques for noise reduction are constantly improved to enable lower radiation doses in CT without loss in image quality. The denoising can be realized by machine learning, especially by denoising autoencoders. We developed and tested a variety of network structures and optimization procedures, i.e., learning rules, to remove noise from CT im-ages, to improve the image quality. We present a promising approach for this task.

Methods

A variety of multiple parallel hidden layers denoising autoencoder (MPHL-DEA) structures and optimization proce-dures, e.g. L2 loss function, were used to assess their denoising performance on clinical CT scans. The training was conducted on a dataset containing seven individual CT scans of different patients, while two separate CT scans were used to evaluate the performance. A software provided by Siemens (CTRecon) was used to reconstruct low dose simulations of these scans to provide noisy samples. This enables a training and a testing of various MPHL-DEAs’ denoising performances by different measures.

Results

All MPHL-DEA showed a decent denoising performance. A big gap in efficiency and denoising quality between differ-ent network structures and training/network parameters was observed. Additionally, we found that denoising capa-bilities depend also on the optimization procedure. However, the classical L2 loss function often led to blurred im-ages, which could be avoided by using the structural similarity index as a loss function. By optimizing the learning rule a SSIM improvement of 5-10% was reached.

Conclusion

Considering the low number of data samples used for training and testing, the quality of the results is acceptable up to good. As with most tasks in the field of machine learning or deep learning, not only the optimization of the net-work design, but also an optimization of the learning data, e.g. larger number of available data, would be required for a better denoising of CT images acquired at low dose.

The analysis of spontaneous mind wandering is crucial for comprehending an individual mental state and holds the potential to enhance performance and productivity. This paper proposes a framework using Continuous Wavelet Transform (CWT) based ResNet model to analyze Electrodermal Activity (EDA) signals for mind wandering detection. In this analysis, EDA signals are sourced from an openly accessible database and preprocessed for artifact and noise removal. Time-frequency analysis generates CWT spectrogram images, which are classified using a modified ResNet50 model that is custom built to classify the spectrogram images corresponding the mind wandering and awareness. Hyperparameter tuning is carried to obtain the optimal network parameters that provides the best accuracy. Results indicate that batch size of 32, learning rate 1e-5 provides better results. This hyperparameter tuned model achieved an accuracy of 64% in differentiating between the two classes. This paper proposes an adapted ResNet50 model that could be employed in wearable devices as a potential application of knowing the mind awareness of an individual.

Contactless measurement methods offer a novel approach to assessing respiratory parameters. This study investigates the feasibility of classifying chronic obstructive pulmonary disease, asthma, and healthy individuals using depth-based plethysmography (DPG). The approach involves calculating Pearson's correlation coefficient for all pixel-wise signals against each other, with the cumulative result visualized in patient-specific masks. A convolutional neural network is used for the classification process. For evaluation, on a recorded data set (N=53), a classification accuracy of 57.7 % and Cohen’s Kappa of 0.28 were reached. These findings suggest that DPG can effectively classify respiratory conditions by analyzing respiratory motion dynamics.

Introduction

In addition to sector-specific regulations, e.g. Medical Devices Regulation (MDR), the new European Artificial Intelligence Act (AIA) is applicable for medical devices based on Artificial Intelligence (AI-MD). One of the most challenging AIA requirements is the transparency of AI systems that “relates to making the data, features, algo-rithms, training methods and quality assurance processes available to external inspection by a stakeholder” (ISO/IEC TR 24028). Essentially the following provisions concerning transparency are laid down in the AIA:

• Technical documentation of AI systems shall contain a “detailed description of the elements of the AI system and of the process for its development” as well as “detailed information about the monitoring, functioning and control of the AI system” (Art. 11 and Annex IV)

• AI systems must be designed and developed to ensure that their operation is sufficiently transparent to allow users to interpret the output of the system and to use it appropriately (Art. 13 (1))

• AI systems shall be accompanied by “concise, complete, correct and clear” IFU being “relevant, accessi-ble and comprehensible to users” (Art. 13 (2))

• Natural persons interacting directly with AI systems shall be aware of this interaction (Art. 52)

In the following paragraphs a practical approach is presented for meeting AIA transparency requirements.

Methods

First, relevant stakeholders in the healthcare sector were identified who have a legitimate interest in the transpar-ency of AI-MD. Second, regulatory documents dealing with transparency of AI systems, such as guidelines and standards, were identified and analysed. Third, a series of documents were developed specifically tailored to the needs of potential stakeholders, considering their level of education and knowledge.

Results

In the healthcare sector user (e.g., physician), patient, competent authority, notified body, and healthcare provid-ers (e.g., clinic) are the relevant stakeholders having an interest in the transparency of AI-MD. Examples of regu-latory documents discussing transparency are the ITU publication “Good practices for health applications of ma-chine learning: Considerations for manufacturers and regulators” as well as standards developed by the ISO/IEC JTC 1/SC 42.

As part of the technical documentation of AI-MD the following documents shall be prepared by the manufacturer during the application of respective processes: AI model development plan and report, data management report, and AI model evaluation plan and report. These documents are dedicated to the competent authority and notified body that involved in the market surveillance and the conformity assessment procedure of the AI-MD.

In addition, the AI model technical information and transparency information as part of the instructions for use (IFU) ensure the necessary transparency towards user, patient, and healthcare provider. Whereas the technical information contains technical characteristics, clinical benefits, and residual risks of the AI-MD, the transparen-cy information provides in clear language information regarding general purpose and functioning, rational for us-ing AI technology, safety and performance, currentness, general risks, official product certifications, and limita-tions.

Conclusion

Here it was shown how transparency for AI-MD can be implemented while considering the different interests of stakeholders in the healthcare system. Moreover, this approach is also suitable for markets outside the EU, since regulators world-wide apply common principles, and it is promoting the use of innovative and trustworthy AI in clinical practice.

Introduction

Locating biopsies and mapping tumors in the urinary bladder requires accurate orientation within the hollow organ de-spite the limited field of view in endoscopic setups. To address this need, a digital representation based on monocular endoscopic camera videos is proposed to enable precise localization and storage of all collected data. However, due to the intraoperative setting, the generation of the digital replica faces challenges such as varying brightness, reflections, and continuous underlying deformations.

Methods

Localization relies on accurate feature extraction and mapping. Due to the intraoperative challenges, landmark de-scriptions based on adjacent pixel features are insufficient. Therefore, neural network approaches for feature extrac-tion have been employed in the literature. Here, detection and definition occur simultaneously in a single forward pass. A convolutional neural network (CNN) is employed on input images, yielding a 3D tensor as feature maps with several channels. Descriptors are then obtained by slicing through all channels and, for detection, scored by an algorithm that quantifies their significance. These descriptors are compared for registration using Euclidean distance. However, this concept for feature extraction must be refined for deformable environments. To this end, training data has to be gener-ated to simulate deformation with which the CNN can be trained to extract reliable features under deformation. The created dataset includes synthetic transformed endoscopic images paired with corresponding matching positions.

Results

The simultaneous detection and description exceeds pixel-based methods in feature detection for intraoperative envi-ronments. Moreover, by incorporating a RANSAC filter, reliable matching becomes achievable even in deformed blad-der regions lacking intuitively observable properties.

Conclusions

The introduced feature description method could improve bladder cancer diagnosis and treatment planning through digital bladder modeling. With simultaneous detection and description, refining the matching process and the adapta-tion of the RANSAC filter to deformable environments, the intraoperative challenges of the orientation can be over-come.

A novel two-phase aerosolisation mode utilizing porous materials is investigated, aiming to improve aerosol delivery for medical inhalation. Sintered stainless steel filters with varied pore sizes (PS) from 0.2 μm to 7 μm were used to generate aerosols from a 0.9 wt.% sodium chloride solution. Droplet sizes and delivery rates were measured using laser diffraction spectroscopy. Further measurements included shadow imaging. Results indicate that aerosolisation occurs within a specific range of PS with droplet sizes increasing with increasing PS. The droplets generated are suitable for inhalation therapies. A hypothesis is established about the process of droplet formation which states that different PS within the porous material serve distinct functions that contribute to the breakup of liquid films into aerosol particles. Droplet formation is the result of film breakup in pores filled with fluid. This low-energy aerosolisation method has the potential to be used in handheld devices for sensitive drug formulations, overcoming the limitations of current technologies. Further research is needed to optimize the pore size distribution and enhance aerosol generation efficiency.

Introduction

Transverse Intrafascicular Multichannel Electrodes (TIME) is a penetrating electrode array used to stimulate multiple axon bundles within the nerve. This provides higher coverage and local selectivity, which translates into finer motor control, when used in neurorehabilitation. However, the high number of electrodes accompanies the need for a high-density feedthrough to fit into the limited space of an implant.

Traditional feedthrough fabrication techniques pose a limit to miniaturization in order to maintain the hermet-ic structure and electrical isolation between channels. This restricts the density of the feedthrough, limiting either the quantity of connected electrodes or the degree of miniaturization.

Methods

A novel three-step feedthrough fabrication technique has been devised. Step one involves creating a photo-resist mold for the matrix of dense tracks, using ultra-high resolution, two-photon 3D printing. Step two, this mold is embedded in a ceramic slurry comprised of Hydroxyapatite nanoparticles (nHA). The assembly is sintered such that the ceramic enters a ‘green’ state, which holds shape while the 3D printed structure is burned out, leaving an intricate array of channels. In step 3, these channels are backfilled with a metal paste, and again sintered to achieve a fully dense ceramic filled with electrically isolated and conductive channels.

Results

The conceptualised approach exploits emerging technologies, which open fabrication pathways not previously available. Our preliminary results indicate that co-firing the nHA ceramic slurry and photo-resist mold forms a corresponding network of channels embedded within the ceramic. Current work aims to optimize the sintering characteristics, to ensure the resultant structure is electrically conductive and fully dense, to achieve a functional and reliable feed-through.

Conclusion

As two of the major drivers of implantable technology development are to reduce invasiveness through miniaturization, and increase impact by increasing number of electrical interfaces, this optimization lays the foundation to achieve these goals through the improvement of high-density feedthroughs.

Anatomy-based fitting (ABF) has the potential to lead to better outcomes than conventional clinically-based fitting (CBF). It has already been shown that anatomy-based fitting based on postoperative imaging (postopABF) can improve speech understanding. Moreover, it was demonstrated that patients preferred anatomy-based fitting maps based on preoperative imaging (preop ABF) over CBF. The goal of this study was to examine if a difference between preopABF and postopABF could be observed. Via the software OTOPLAN4 (CAScination, MED-EL) the electrode position was calculated based on preoperatively, as part of the clinical routine, obtained imaging (CT or MRI). If necessary, after the cochlear implantation surgery the selected electrode array was adapted if a different array was chosen intraoperatively. This data was exported into the fitting software MAESTRO to enable a preopABF. The same process was done based on postoperative imaging (CT) so that 1) the frequency bands were compared between preopABF and postopABF data and 2) the difference was displayed in semitones. 27 Patients could be included who got a postoperative CT scan. In addition, subjective tinnitus and dizziness or imbalance issues did not significantly change from the pre-operative appointment over the first fitting to the control fitting one month later. In this study, we could show that there is no significant difference between apical to medial and a significant difference for medial to basal electrode contacts between the preopABF and postopABF regarding the frequency bands. Thus, preopABF is an easy, clinically practicable way to implement ABF, which is promising to lead to better outcomes without extra costs and the radiation burden that comes with postopABF. If patients subjectively prefer either preopABF or postopABF is still to be examined.

Introduction: In auditory evoked potentials (= AEP), auditory brainstem response (= ABR) is the gold standard for objective analysis of the auditory pathway using acoustical stimulators like headphones during hearing screening and patient assessment. Where no response can be recorded, electrical stimulation can be used in pre-, intra-, or postoperative scenarios to confirm the excitability of the auditory nerve (= eAEP). In eAEP, the recording of brainstem response (= eABR) is mostly the method of choice. Sometimes in eABR, the stimulation artifact can not only harden but even prevent the detection of the response waveforms. Methods: Many factors can influence the stimulation artifact when recording an eAEP. Neurological recording systems are recommended as well as some approved AEP systems traditionally used in ENT for amplification to avoid large stimulation artifacts in the response waveforms, but able to recover in <1 ms from saturation. High stimulation rates (~90 Hz) can lead to overlapping of the response waveforms with the following stimulation artifact. In preoperative eAEP, stimulation artifacts are often high due to extra-cochlear stimulation. In intra- and postoperative eAEP, the artifact is lower and mostly similar across patients. Results: The recommended recording systems show low stimulation artifacts in almost any case for intra- and postoperative eABR. In contrast, later responses, like eAMLR and eALR, are less affected by artifacts. Conclusion: The stimulation artifact can be a problematic factor. Depending on the EP system, intra- and postoperative eABR can be challenging when trying to record the response waves. Where possible, clinicians should think about eAMLR or eALR for results with less stimulation artifact. Finally, the most important factor remains using a recording system that can handle the stimulation artifact. We do not recommend recording devices with high artifacts for preoperative eABR. For intra- and postoperative eABR, the recommended systems can prevent challenging recording.