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

Abstract

Recording open surgery procedures is essential for educational and clinical evaluation purposes; however, traditional single-camera methods often face challenges such as occlusions caused by the surgeon's head and body, as well as limitations due to fixed camera angles, which undermine the comprehensibility of the recorded surgical content. In this study, we specifically focus on open thyroidectomy and employ a multi-viewpoint camera recording setup, in which six synchronized cameras capture the surgery from different angles simultaneously. We develop a supervised time-series prediction framework to automatically select the most informative camera views, ensuring better coverage of critical steps. Our model forecasts camera selections by extracting and fusing visual and semantic features from thyroidectomy videos using pre-trained models, followed by temporal modeling with TimeBlocks. We constructed a dataset of five thyroidectomy procedures with synchronized six-view recordings and conducted experiments. The results show that our method achieves stable accuracy compared with existing baselines and outperforms several mainstream time-series prediction models in this specific surgical scenario. This work provides an initial exploration of multi-view camera selection for thyroidectomy, with potential value for surgical video documentation and training. Code is available at https://github.com/Aveouter/SurgicalCamSwitch.