The interpretable surgical temporal informer: explainable surgical time completion prediction.

Area of Science:

• Artificial Intelligence in Medicine
• Surgical Workflow Optimization
• Medical Imaging Analysis

Background:

• Accurate surgical time prediction enhances operating room (OR) utilization and hospital efficiency.
• Surgeon technical proficiency is a key factor influencing surgical completion time.
• Interventional videos offer rich data for analyzing surgical performance and predicting duration.

Purpose of the Study:

• To develop an interpretable method for predicting surgical duration using egocentric surgical videos.
• To identify visual features within surgical videos that correlate with time prediction accuracy.
• To enhance the understanding of factors influencing surgical completion times.

Main Methods:

• Introduced an interpretable AI model identifying prototype-like patterns in surgical videos.
• Extracted video-based explanations linked to time deviation patterns, moving beyond conventional image patches.
• Utilized an informer model to capture long-range dependencies for accurate duration prediction.

Main Results:

• The interpretable model was applied to 42 craniotomy videos.
• The proposed method demonstrated superior performance compared to baseline models in predicting surgical time completion.
• The model provides prototype-like visual explanations for its predictions.

Conclusions:

• The approach enhances the interpretability of surgical time predictions.
• Leverages detailed information from surgical video data for improved accuracy.
• Offers a novel method for analyzing surgical video to predict duration and potentially assess skill.