Related Experiment Video
Updated: Aug 31, 2025
08:27
Three-Dimensional Finger Motion Tracking during Needling: A Solution for the Kinematic Analysis of Acupuncture Manipulation
Published on: October 28, 2021
2.9K
Developing artificial intelligence models for medical student suturing and knot-tying video-based assessment and
Madhuri B Nagaraj1,2, Babak Namazi3, Ganesh Sankaranarayanan3
1Department of Surgery, University of Texas Southwestern Medical Center, 5323 Harry Hines Blvd, Dallas, TX, 75390-9159, USA. Madhuri.nagaraj@gmail.com.
Surgical Endoscopy
|August 18, 2022
Summary
An AI model was developed to assess medical students' suturing skills via video, achieving high accuracy in identifying instrument-holding errors. This technology offers automated feedback for surgical training, addressing limitations in traditional assessment methods.
Area of Science:
- Medical Education
- Surgical Skills Training
- Artificial Intelligence in Healthcare
Background:
- Suturing skills are crucial for medical students, but traditional training faces challenges with feedback frequency and directedness.
- The COVID-19 pandemic necessitated adaptation of in-person surgical skills curricula to remote, video-based assessments.
- Developing automated assessment tools is essential for efficient and effective surgical skill acquisition.
Purpose of the Study:
- To develop and validate an Artificial Intelligence (AI) model for automated video-based assessment of basic suturing skills.
- To evaluate the AI model's ability to identify specific errors in instrument handling and knot tying during surgical procedures.
- To provide a scalable solution for delivering timely and targeted feedback to medical students.
Main Methods:
- Second-year medical students self-trained and submitted videos of performing a simple interrupted knot with instrument tying.
- Two AI models utilizing convolutional neural networks were developed and trained to detect instrument-holding and knot-tying errors.
- A k-fold cross-validation (k=10) was employed to assess model performance on a dataset of 216 videos.
Main Results:
- The instrument-holding AI model achieved 89% accuracy with a 74% F-1 score.
- The knot-tying AI model demonstrated 91% accuracy with a 54% F-1 score.
- Analysis of 229 videos revealed common errors in instrument-holding (47) and knot-tying (15).
Conclusions:
- AI-powered automated surgical video analysis can effectively supplement traditional assessment methods for surgical skills.
- The developed AI model shows promise in providing objective and efficient feedback, addressing limitations in time-consuming manual assessments.
- Future work will focus on refining the AI model to identify discrete errors and enhance the specificity of feedback for surgical trainees.
