Related Experiment Video
Updated: Jul 19, 2025
A Teleoperated Robotic System-Assisted Percutaneous Transiliac-Transsacral Screw Fixation Technique
Published on: January 6, 2023
Artificial Intelligence for context-aware surgical guidance in complex robot-assisted oncological procedures: An
Fiona R Kolbinger1, Sebastian Bodenstedt2, Matthias Carstens3
1Department of Visceral, Thoracic and Vascular Surgery, University Hospital and Faculty of Medicine Carl Gustav Carus, Technische Universität Dresden, Fetscherstraße 74, 01307 Dresden, Germany; National Center for Tumor Diseases Dresden (NCT/UCC), Germany: German Cancer Research Center (DKFZ), Heidelberg, Germany; Faculty of Medicine and University Hospital Carl Gustav Carus, Technische Universität Dresden, Dresden, Germany; Helmholtz-Zentrum Dresden - Rossendorf, Dresden, Germany; Else Kröner Fresenius Center for Digital Health (EKFZ), Technische Universität Dresden, Fetscherstraße 74, 01307, Dresden, Germany.
Machine learning models can recognize surgical phases and segment structures during robot-assisted rectal resection (RARR). This technology shows promise for improving surgical guidance and patient outcomes in complex rectal surgeries.
Area of Science:
- Robotics and Artificial Intelligence in Surgery
- Surgical Oncology
- Medical Imaging Analysis
Background:
- Robot-assisted rectal resection (RARR) presents challenges in maintaining dissection planes and protecting nerves, crucial for preventing local recurrence, incontinence, and sexual dysfunction.
- Accurate identification of surgical phases and anatomical structures is vital for successful RARR outcomes.
Purpose of the Study:
- To investigate the feasibility of using machine learning for surgical phase recognition and target structure segmentation in RARR.
- To evaluate the performance of different machine learning models in these tasks.
Main Methods:
- Trained three machine learning models (LSTM, MSTCN, Trans-SVNet) for surgical phase recognition on 57 recorded RARR procedures.
- Developed pixel-wise segmentation models (DeepLabv3) for anatomical structures, tissue types, and dissection areas using 9037 annotated images.
- Evaluated model performance using metrics like F1 score, Intersection-over-Union (IoU), accuracy, precision, recall, and specificity.
Main Results:
- The MSTCN model achieved the best performance in phase recognition (F1 score: 0.82 ± 0.01, accuracy: 0.84 ± 0.03).
- Mean IoUs for target structure segmentation varied, with organs/tissue types ranging from 0.14 ± 0.22 to 0.80 ± 0.14, and dissection areas from 0.11 ± 0.11 to 0.44 ± 0.30.
- Image quality and intraoperative factors like blood and smoke significantly affected segmentation accuracy.
Conclusions:
- Machine learning-based surgical phase recognition and target structure segmentation are feasible in RARR.
- These AI-driven functionalities hold potential for integration into future context-aware surgical guidance systems for rectal surgery.
