TSP-OCS: A Time-Series Prediction for Optimal Camera Selection in Multi-Viewpoint Surgical Video Analysis

Area of Science:

• Medical Imaging
• Computer Vision
• Surgical Education

Background:

• Traditional single-camera surgical recordings suffer from occlusions and fixed angles, limiting educational and clinical evaluation value.
• Open thyroidectomy documentation requires comprehensive visual data for effective training and assessment.

Purpose of the Study:

• To develop an automated system for selecting optimal camera views during open thyroidectomy.
• To enhance the comprehensibility and utility of surgical video recordings for training and evaluation.

Main Methods:

• Implemented a multi-viewpoint synchronized camera setup (six cameras) for recording open thyroidectomy.
• Developed a supervised time-series prediction framework using visual and semantic features with TimeBlocks for camera view selection.
• Constructed a dataset from five synchronized six-view thyroidectomy procedures.

Main Results:

• The proposed method demonstrated stable accuracy compared to existing baselines.
• The system outperformed several mainstream time-series prediction models in the context of surgical video analysis.
• Successfully automated the selection of informative camera views for critical surgical steps.

Conclusions:

• Multi-view camera selection using AI offers a valuable approach to improve surgical video documentation for thyroidectomy.
• This technology has significant potential for enhancing surgical training and clinical evaluation.
• The developed framework provides a foundation for future research in automated surgical video analysis.