Falls in the elderly population present substantial health risks, often resulting in morbidity and a decline in quality of life. Conventional fall detection methods, including wearable devices and cameras, are hindered by issues such as variable lighting conditions and privacy considerations. Radar-based fall detection has emerged as a promising alternative, providing an unobtrusive method. This study aims to classify fall detection using Smoothed Pseudo Wigner-Ville distribution (SPWVD) images and ResNet-50 model. For this, an online publicly available radar database comprising 15 subjects is utilized. Radar signals is processed into SPWVD time-frequency representation images for analysis. The SPWVD images is fed into the ResNet-50 model. Experiments are performed and performance is evaluated using 10- fold cross validation. The proposed approach is able to distinguish elderly fall. Using ResNet-50 model, the approach yields a maximum average classification f-measure, area under curve score, precision and, sensitivity of 67.15%, 52.31%, 50.55% and, 99.88%, respectively. Hence, the proposed framework holds promise for accurately and efficiently detecting falls among the elderly population within their private environments.

Nursing care jobs are predominantly described as

burdensome. In particular, the workload and the time available

to deal with it are in a state of imbalance. Nurses are very often

under time pressure and have to make decisions under psychological,

emotional and physical stresses. Machine learning

methods and in particular, Decision Support Systems (DSS)

can be used to support nurses with improved information for

decisions that are not easy to make. As falls are one of the most

common health problems in care facilities, we will present a

concept of a DSS for the prevention of falls that was developed

in the PFLIP research project. We explain our user-centered

design process based on ISO 9241-210-2019. User feedback

is obtained at each phase so that nurses can identify problems

at an early stage. The result is a conceptual design, visualized

as a click dummy, to identify individual fall risks and predict

preventive measures to reduce the risk of falling.

Introduction

The combination of physiological parameters in medical practice enables a more comprehensive understanding of the state of health and contributes to early diagnosis[1, 2]. Advances in microelectronics are leading to the development of new sensors that allow the creation of new sensor combinations.

For this reason, a measuring platform has been set up to test various sensors, and to perform series of physiological measurements. The focus is on the time-synchronous recording and real-time visualization of acquired multisensor data. Simultaneously, another goal is to create a versatile platform for versatile sensor technologies, enabling the integration of different sensors to address diverse medical concerns. By utilizing specialized algorithms, we aim to analyse the data for individual health patterns and uncover novel findings in psychological medicine.

Methods

As a first iteration of the measuring platform, an available plug-in system with various boards for capturing different biometric signals like electrocardiogram (ECG), an electrooculogram (EOG) and the respiratory activity is used (supplier: MicroE). A special firmware that was written to perform the real-time recording of up to 6 sensors, the filtering of interference signals and the transmittance of the data to a server. The server enables interactive real-time visualization and calculates relevant metrics such as heart rate, respiratory rate, number of saccades and blinks caused by eye movements online. After recording, further statistical analysis and visualization are available, such as heart rate variability with its frequency spectrum and Poincaré plot. All recordings are organized in a customized database.

Results

Series of measurements for ECG, EOG, and the respiratory curve were conducted. The determined medical metrics and the visual analysis of the recorded data show a typical characteristic as described in the literature. To confirm the saccades and blinks captured with the EOG accompanying videos were recorded. Additionally, the integration of a skin conductance sensor including the corresponding algorithms is aimed as well as the development of a shielded housing.

Conclusion

As the results showed, our measuring platform enables the time-synchronized recording and real-time visualization of multiple sensors. Based on that, we would like to point out the high potential of multi-sensor technology in the field of Mental Health. In the future, we plan to develop machine learning algorithms for sensor data fusion. Although, we aim to further miniaturize and modularize the measuring platform as well as de-wiring and wireless sensor modules to ensure greater convenience for the patient.

References

[1] Mielke, Corinna (2020): Assistierende Gesundheitstechnologien zum Monitoring von psychischen Erkrankungen am Beispiel der Depression

[2] Ahmed, Tashfia et al. (2022): Physiological monitoring of stress and major depression: A review of the current monitoring techniques and considerations for the future, in: Biomedical Signal Processing And Control, Bd. 75, S. 103591

Breast cancer (BC) poses a significant worldwide health challenge, necessitating the identification of its molecular origins and the development of possible treatment approaches. In this investigation, we introduce a pioneering methodology integrating genomic data and machine learning algorithms to identify genes associated with BC and their corresponding drugs. Initially, RNA-sequencing data of normal and malignant BC tissues publicly available in the NCBI GEO database were pre-processed using a standard pipeline. Further, machine learning algorithms, such as logistic regression, support vector machine, and random forest, were used to identify the differentially expressed genes (DEGs). The results were validated based on accuracy, sensitivity, specificity, precision, and F-score. Moreover, we identified the drugs corresponding to DEGs using the DepMap database. Our results revealed that genes such as OC90, KLK9, CXCL10, CDRT1, LCN6, GOLGA7B, and ZNF223 were commonly identified by all three machine learning algorithms. We found that the drugs DIHYDROROTENONE, 4-IODO-6-PHENYLPYRIMIDINE, BMS-754807, ARRY-886, ESTRAMUSTINE-PHOSPHATE, FR-122047, and PIK 93 produced high correlation values on gene-drug interaction. The present study emphasizes the significance of utilizing genetic data and powerful machine learning algorithms to decipher the intricacies of BC biology and expedite the creation of tailored therapeutic approaches.

Introduction

Large amounts of healthcare data reside in clinical reports and require standardization through medical terminologies such as SNOMED-CT to resolve ambiguity and enable clinical applications, such as heart failure prediction or symptom clustering. However, linking clinical entities in clinical reports, such as mentions of diseases or procedures to clinical concepts is complex: First, it requires clinical domain expertise to understand clinical reports alongside numerous abbreviations and synonyms; Second, knowledge about SNOMED-CT with over 360,000 concepts is needed to accurately link entities. However, the expertise needed for entity linking is expensive and hard to reach. This is even more pronounced for German clinical reports, due to a lack of a comprehensive German SNOMED-CT version and familiarity with the terminology.

Methods

We developed and tested a novel entity linking pipeline for German clinical reports. The pipeline steps entail context-dependent abbreviation disambiguation, German clinical report translation to English through biomedical translation models, and entity linking to SNOMED-CT using a zero-shot semantic-similarity-based approach. An implemented automated check regarding the patient’s demographic and laboratory data determines the plausibility of linked concepts and reduces manual effort when evaluating the results. Based on the results of the plausibility check, linked concepts were sampled and their accuracy was rated on a four-point LIKERT scale in a survey by physicians.

Results

Our pipeline was successfully tested on clinical reports of over 800 heart failure patients from the University Hospital Aachen. Survey results yield sufficient accuracy for correctly linked concepts. The results provide valuable knowledge on adapting the pipeline, for instance by leveraging grammatical structures of clinical reports through dependency parsing, to increase the accuracy.

Conclusion

The results demonstrate the applicability of automatically linking clinical entities in clinical reports to SNOMED-CT and open pathways for future applications, such as heart failure symptom clustering and heart failure prediction in patients.

This paper examines the challenges and opportunities associated with the implementation of Electronic Health Records (EHR) globally and, specifically, in Germany. While EHRs promise improved accessibility and comprehensibility of patient data, healthcare professionals face challenges such as resource constraints and the need for user-friendly interfaces. The focus is on the design of the graphical user interface (GUI) to enhance accessibility for both healthcare professionals and patients. The proposal suggests a three-dimensional digital human model as the basis for organizing and retrieving medical documents, integrating visual and chronological aspects. Additionally, the paper explores the potential benefits of incorporating personal health data, addressing data access separation concerns for users and promoting interdisciplinary information exchange. The concept-study concludes by highlighting the importance of a well-designed user interface in facilitating effective utilization of EHR systems.