Automated laparoscopic video analysis is essential for assisting surgeons during computer aided medical procedures. Nevertheless, it faces challenges due to complex surgical scenes and limited annotated data. Most of the existing methods for classifying surgical tools in laparoscopic surgeries rely on conventional deep learning methods such as convolutional and recurrent neural networks. This paper explores the use of pure self-attention based models—Vision Transformers for classifying both single-label (SL) and multi-label (ML) frames in Laparoscopic surgeries. The proposed SL and ML models were comprehensively evaluated on the Cholec80 surgical workflow dataset using 5-fold cross validation. Experimental results showed an excellent classification performance with a mean average precision mAP=95.8% that outperforms conventional deep learning multi-label models developed in previous studies. Our results open new avenues for further research on the use of deep transformer models for surgical tool detection in modern operating theaters.

Vibroacoustic signals generated by the interaction of moving clinical devices, e.g. an aspiration or biopsy needle, with different tissues creates a distinct signal. These signals can be received via a dedicated audio sensor and subsequently analysed with the potential to provide information about location, tissue characterisation, and event classification. With that it could also be used as an additional and complementary guidance tool particularly for future robotic assisted procedures. In our laboratory research we used different phantoms with animal and artificial tissues, audio pre-processing, and subsequent training and optimisation of a ResNet model with a tissue event detection F1 score of 95.6% when compared to video based annotation tool. This result is very encouraging, as several possible improvements have been identified that will be implemented in the next research steps together with a robot assisted insertion and an automatic video annotation algorithm.

Beating-heart surgery is performed to reduce patient trauma but depending on the intervention requires tissue stabilization through vacuum cardiac tissue stabilizers. Current designs either require open chest sternotomy or display significant residual motion when used minimally invasively. Active motion compensation methods require complex control algorithms and expensive technology. We propose a novel design for a vacuum tissue stabilizer with expandable suction feet enabling a larger stabilized area and increased motion reduction. Its performance is evaluated on a heart phantom using a silicon membrane mimicking cardiac tissue properties. The motion of the membrane is captured using a stereo camera and marker points on the membrane surface. The results are then compared to a state-of-the-art tissue stabilizer. The extended stabilizer is found to stabilize a 110% larger area while achieving similar motion amplitude reduction. Thus, extendable stabilizers seem to be a promising solution for improved cardiac tissue stabilization in less invasive beating-heart surgery.

Robotic systems have the potential to improve the accuracy and efficiency of percutaneous interventions. However, their widespread adoption is hindered by challenges such as high costs and usability deficiencies. To address these challenges, we designed a novel semi-automated robotic system that combines manual and robotic degrees of freedom for needle guidance. Since the initial presentation of the system, we have implemented hardware and software improvements. The objective of this study is to evaluate the accuracy of the system. To achieve this, experiments were conducted that involved the entire process from CT scanning to lesion targeting with a needle, using a phantom that replicated abdominal lesions of varying sizes. A total of 36 needle insertions were performed, and a control scan was conducted each time to evaluate the needle tip position. The results demonstrate promising accuracy, with a mean Euclidean error of 5.8 ± 1.7 mm. Qualitative analysis showed that lesion targeting was successful across different sizes, with a total success rate of 27 out of 36 attempts. Although our prototype shows promising accuracy, there is still room for improvement compared to existing systems. Future efforts will prioritize enhancing accuracy, exploring usability aspects, and validating the system's performance with medical professionals