Surgical workflow analysis optimizes efficiency, resource use, and patient safety in catheterization labs. Traditional manual methods are labour-intensive and inconsistent, driving the need for automated solutions that utilize machine learning and computer vision. This thesis introduces an explainable two-stage model for workflow analysis using ceiling-mounted cameras. The approach combines a YOLOv8 object detection model with a Gaussian Mixture Model - Hidden Markov Model (GMM-HMM). The first stage detects key objects for input into the second stage, where the GMM-HMM infers workflow phases by modelling spatial and temporal dynamics for real-time classification. Validation on two hospital datasets achieves 95.2% accuracy for the RdGG dataset and 95.4% for HH Tampere, demonstrating generalizability across environments. Experimental results show high accuracy in detecting workflow phases, highlighting explainability and robustness. The combined efficiencies of YOLOv8 and GMM-HMM allow for precise phase transition identification. The model's real-time application and adaptability across hospitals suggest its clinical implementation potential. This research furthers automated workflow analysis by enhancing interpretability and adaptability. Future work aims to improve robustness against occlusions, integrate audio data, and explore applications in other surgical settings.

This study evaluates the performance of deep learning models in the prediction of the end time of procedures performed in the cardiac catheterization laboratory (cath lab). We employed only the clinical phases derived from video analysis as input to the algorithms. Our results show that InceptionTime and LSTM-FCN yielded the most accurate predictions. InceptionTime achieves Mean Absolute Error (MAE) values below 5 min and Symmetric Mean Absolute Percentage Error (SMAPE) under 6% at 60-s sampling intervals. In contrast, LSTM with attention mechanism and standard LSTM models have higher error rates, indicating challenges in handling both long-term and short-term dependencies. CNN-based models, especially InceptionTime, excel at feature extraction across different scales, making them effective for time-series predictions. We also analyzed training and testing times. CNN models, despite higher computational costs, significantly reduce prediction errors. The Transformer model has the fastest inference time, making it ideal for real-time applications. An ensemble model derived by averaging the two best performing algorithms reported low MAE and SMAPE, although needing longer training. Future research should validate these findings across different procedural contexts and explore ways to optimize training times without losing accuracy. Integrating these models into clinical scheduling systems could improve efficiency in cath labs. Our research demonstrates that the models we implemented can form the basis of an automated tool, which predicts the optimal time to call the next patient with an average error of approximately 30 s. These findings show the effectiveness of deep learning models, especially CNN-based architectures, in accurately predicting procedure end times.

Background: The objective of an operating room (OR) ultra-clean ventilation system is to eliminate or reduce the quantity of dust particles and colony-forming units per cubic meter of air (CFU/m³). To achieve this, ultra-clean goal high air change rates per hour are required to reduce the particle load and number of CFU/m³. Aim: To determine the air quality in an ultra-clean OR during surgery, in terms of the number and type of microorganism and quantity of dust particles in order to establish a benchmark. Methods: Number of CFUs and the quantity of dust particles were measured. For measuring the CFUs, sterile extraction hoses were positioned at the incision, the furthest away positioned instrument table, and the periphery. At these locations, air was extracted to determine the quantity of dust particles. Findings: The number of CFU/m³ and particles was on average at wound level ≤1 CFU/m³ resp. 852.679 particles, at instrument table ≤1 CFU/m³ resp. 3.797 particles and in the periphery ≤8 CFU/m³, resp. 4.355 particles. Conclusion: The number of CFUs in the ultra-clean area is below the defined ultra-clean level of ≤10 CFU/m³ for ultra-clean surgery. The quantity of dust particles measured during surgery was higher than the defined ISO 5.

Collision feedback about instrument and environment interaction is often lacking in robotic surgery training devices. The PoLaRS virtual reality simulator is a newly developed desk trainer that overcomes drawbacks of existing robot trainers for advanced laparoscopy. This study aimed to assess the effect of haptic and visual feedback during training on the performance of a robotic surgical task. Robotic surgery-naïve participants were randomized and equally divided into two training groups: Haptic and Visual Feedback (HVF) and No Haptic and Visual Feedback. Participants performed two basic virtual reality training tasks on the PoLaRS system as a pre- and post-test. The measurement parameters Time, Tip-to-tip distance, Path length Left/Right and Collisions Left/Right were used to analyze the learning curves and statistically compare the pre- and post-tests performances. In total, 198 trials performed by 22 participants were included. The visual and haptic feedback did not negatively influence the time to complete the tasks. Although no improvement in skill was observed between pre- and post-tests, the mean rank of the number of collisions of the right grasper (dominant hand) was significantly lower in the HVF feedback group during the second post-test (Mean Rank = 8.73 versus Mean Rank = 14.27, U = 30.00, p = 0.045). Haptic and visual feedback during the training on the PoLaRS system resulted in fewer instrument collisions. These results warrant the introduction of haptic feedback in subjects with no experience in robotic surgery. The PoLaRS system can be utilized to remotely optimize instrument handling before commencing robotic surgery in the operating room.

Deep learning-based object detectors, while offering exceptional performance, are data-dependent and can suffer from generalization issues. In this work, we investigated deep neural networks for detecting people and medical instruments for the vision-based workflow analysis system inside Catheterization Laboratories (Cath Labs). The central problem explored in this paper is the fact that the performance of the detector can degrade drastically if it is trained and tested on data from different Cath Labs. Our research aimed to investigate the underlying causes of this specific performance degradation and find solutions to mitigate this issue. We employed the YOLOv8 object detector and created datasets from clinical procedures recorded at Reinier de Graaf Hospital (RdGG) and Philips Best Campus, supplemented with publicly accessible images. Through a series of experiments complemented by data visualization, we discovered that the performance degradation primarily stems from data distribution shifts in the feature space. Notably, the object detector trained on non-sensitive online images can generalize to unseen Cath Labs, outperforming the model trained on a procedure recording from a different Cath Lab. The detector trained on the online images achieved an mAP@0.5 of 0.517 on the RdGG dataset. Furthermore, by switching to the most suitable camera for each object in the Cath Lab, the multi-camera system can further improve the detection performance significantly. An aggregated L-camera mAP@0.5 of 0.679 is achieved for single-object classes on the RdGG dataset.

Background: Surgical instrument counting is a manual, attention-intensive task of the operating room (OR) nurse. Many labour-saving technologies have been proposed, but implementation remains challenging. Knowledge of current counting methods and staff preferences could guide future developments towards effective application. Approach: We observed OR nurses counting materials and instruments in 50 surgical procedures performed by various surgical specialties in a regional teaching hospital in Delft, The Netherlands. Additionally, we surveyed them on their preferences concerning the methods of counting. Key findings: Variations in approaches of surgical counting were observed, with OR nurses using multiple strategies and counting techniques to manage disruptions and limit workload. Interest in using supportive technology is limited to the preoperative and postoperative phase. Relevance: This research relates observational data to staff preferences. Our findings may guide future developments of labour-saving innovations regarding surgical counting towards developing more effective applications and to ensure successful implementation.

When making a decision on the operating room air handling installation and type of air supply system, it is relevant to know the expenditures of the different air handling installations and air supply systems. The aim of this study was to determine the capital and operational expenditures of air handling installations equipped with an ultra-clean or with a conventional system. To compare the technical requirements of Dutch air handling installations with European standards and guidelines, and evaluate the costs of surgical site infections in comparison with the capital expenditures. This study fills a gap in knowledge, detailed technical information and costs of air handling installations and air supply systems from multiple completed projects of 24 hospitals were collected, analyzed and compared. Per OR capital expenditures increase by €62,491 to €139,018 when an air handling installation with an ultra-clean system is compared to a conventional system, which is 3%–7% of the total construction costs of a completely new OR department. The yearly increase in operational expenditures per OR with an ultra-clean system compared to a conventional system was €673 to €1,896. The capital and operational expenditures of air handling installations with an ultra-clean system are higher than those with a conventional system. The technical specifications of the ORs studied in the Netherlands correspond to European standards and guidelines. When the impact on patient suffering and costs associated with surgical site infections are weighed against the investment required for an air handling installation with an ultra-clean system, it is worth considering.

The authors regret that they have missed during the review process that the legend belonging to figure 6a, 6b, 6c and 6d is not shown. It is necessary to show the legenda otherwise the readers cannot interpret the graphs. Apologies that we missed it during the review process of the article. The legend belonging to figure 6 is:[Formula presented] The authors would like to apologize for any inconvenience caused.

Background: As global use of surgical robotic systems is steadily increasing, surgical simulation can be an excellent way for robotic surgeons to acquire and retain their skills in a safe environment. To address the need for training in less wealthy parts of the world, an affordable surgical robot simulator (PoLaRS) was designed. Methods: The aim of this pilot study is to compare learning curve data of the PoLaRS prototype with those of Intuitive Surgical’s da Vinci Skills Simulator (dVSS) and to establish face- and construct validity. Medical students were divided into two groups; the test group (n = 18) performing tasks on PoLaRS and dVSS, and the control group (n = 20) only performing tasks on the dVSS. The performance parameters were Time, Path length, and the number of collisions. Afterwards, the test group participants filled in a questionnaire regarding both systems. Results: A total of 528 trials executed by 38 participants were measured and included for analyses. The test group significantly improved in Time, Path Length and Collisions during the PoLaRS test phase (P ≤ 0.028). No differences was found between the test group and the control group in the dVSS performances during the post-test phase. Learning curves showed similar shapes between both systems, and between both groups. Participants recognized the potential benefits of simulation training on the PoLaRS system. Conclusions: Robotic surgical skills improved during training with PoLaRS. This shows the potential of PoLaRS to become an affordable alternative to current surgical robot simulators. Validation with similar tasks and different expert levels is needed before implementing the training system into robotic training curricula.