Related Experiment Video
Updated: Sep 8, 2025
07:46
Author Spotlight: Revolutionizing Remote Surgery with Augmented Reality and Robotics for Enhanced Precision and Accessibility
Published on: August 9, 2024
844
Artificial intelligence-based action recognition and skill assessment in robotic cardiac surgery simulation: a
Gennady V Atroshchenko1,2,3, Lærke Riis Korup4, Nasseh Hashemi5,6
1Department of Cardiothoracic Surgery, Aalborg University Hospital, Hobrovej 18-22, 9000, Aalborg, Denmark. gennady.atroshchenko@gmail.com.
Journal of Robotic Surgery
|July 13, 2025
Summary
This study developed a deep neural network to recognize surgical actions and assess surgeon skill from video. The AI achieved high accuracy in action recognition, showing potential for robotic surgery training.
Area of Science:
- Robotics
- Artificial Intelligence
- Surgical Education
Background:
- Objective assessment of surgical skills is crucial for training and patient safety in robotic surgery.
- Current methods for skill assessment are often subjective and time-consuming.
- AI-powered video analysis offers a potential solution for objective, automated skill evaluation.
Purpose of the Study:
- To develop a deep neural network (DNN) for recognizing surgical actions (suturing, dissection) from video.
- To categorize surgeons as novice or expert based on their performance using video data.
- To establish a foundation for AI-based automated post-procedural assessments in robotic cardiac surgery simulation.
Main Methods:
- Nineteen surgeons of varying experience performed three wet lab tasks on a porcine model.
- A hybrid DNN combining Convolutional Neural Networks (CNN) and Long Short-Term Memory (LSTM) was utilized.
- Temporal labeling was used for action recognition (AR), and recordings were annotated for skill assessment (SA) as novice or expert.
Main Results:
- The action recognition network achieved a mean accuracy of 98% (cross-validation) and 93% (testing), with high precision and recall.
- The skill assessment network achieved 79% accuracy (cross-validation) but requires further data for robust evaluation.
- Gradient-weighted Class Activation Mapping confirmed the AI focused on relevant surgical elements (instruments, tissue).
Conclusions:
- The DNN demonstrated high accuracy in recognizing surgical actions, indicating its potential utility.
- The skill assessment component shows promise but needs more data to become a reliable tool for performance evaluation.
- Combined deep learning models can form the basis for AI-driven automated assessments in robotic surgery training.
