Automated assessment of simulated laparoscopic surgical skill performance using deep learning.

Area of Science:

• Medical Artificial Intelligence
• Surgical Simulation
• Computer Vision in Medicine

Background:

• Artificial intelligence (AI) and computer vision (CV) offer potential for enhancing healthcare and patient safety.
• CV techniques are increasingly applied to analyze surgical videos for training and performance improvement.
• Challenges in surgical AI include the lack of labeled data and the high cost of manual annotation.

Purpose of the Study:

• To introduce the Laparoscopic Surgical Performance Dataset (LSPD), designed for evaluating simulated laparoscopic surgical skill.
• To address the challenge of limited labeled data in surgical training.
• To assess the performance of a 3-dimensional convolutional neural network (3DCNN) in classifying surgeon expertise levels.

Main Methods:

• Collected and utilized the LSPD, a novel dataset for surgical skill evaluation.
• Employed a 3-dimensional convolutional neural network (3DCNN) with a weakly-supervised approach.
• Analyzed surgical simulation videos to compare performance across novice, trainee, and expert skill levels for specific surgical tasks.

Main Results:

• The 3DCNN model effectively distinguished between novice, trainee, and expert surgeons.
• Achieved a high F1 score of 0.91 and an Area Under the Curve (AUC) of 0.92 in classifying surgeon experience.
• Identified specific skills that are poorly and well-executed across different expertise levels.

Conclusions:

• The LSPD dataset is valuable for automated surgical performance evaluation.
• 3DCNN-based, weakly-supervised methods can automate surgical skill assessment, reducing reliance on manual annotation.
• These advancements contribute to improving surgical training and performance analysis.