VasSurDetect: An AI-based Chatbot for Rapid, Photo-based Feedback in Microsurgical Suturing Training

Microsurgical suturing training requires precise, repeated feedback, but faculty feedback is often delayed, subjective, and limited by availability. We developed VasSurDetect, a LINE mobile messaging application (LY Corporation, Tokyo, Japan)-based artificial intelligence chatbot, to provide rapid photo-based feedback on vascular suturing end-product quality.

A total of 350 paired microscope images from simulation-based chicken wing vessel practice were collected as inside and outside views. Knots, purchase segments, and leftover suture material were annotated. Nine object detection models were trained and compared. Bounding-box outputs were converted into quantitative metrics for stitch number, purchase length, interstitch distance, stitch angle, stitch axis, and leftover suture length. The selected model was deployed in a chatbot workflow that accepts 2 microscope photographs and returns an automated report.

For inside-view detection, YOLOv10 achieved the highest overall performance, with an F1 score of approximately 0.89. For outside-view detection, YOLOv8 and YOLOv10 performed comparably, each achieving an F1 score of approximately 0.55. YOLOv10 was selected for system deployment. The implemented workflow generated stitch-level measurements, including stitch spacing, angular alignment, and leftover suture length, with a mean spacing error of approximately 0.4 mm and an angular alignment error of approximately 10 degrees. Automated feedback was delivered within 10 seconds.

VasSurDetect provides rapid, standardized, image-based feedback for microsurgical suturing practice without additional hardware. It is intended to complement, not replace, expert supervision by supporting high-frequency deliberate practice and objective performance tracking in simulation-based microsurgical training.