Surgical Video Machine Learning Platform to Quantify Field of Focus During Posterior Fossa Craniotomies

Introduction: Optimal visualization and field of focus are critical for safe microsurgical technique. Recent advances in artificial intelligence (AI) offer the potential for objective technical assessment from surgical video. We present initial data from a proof-of-concept study using AI to analyze field of focus metrics from intraoperative microsurgical video.

Methods: Operative microscope video from retrosigmoid craniotomies performed by a senior surgeon was analyzed using the Surgical Video Platform (SDSC). A spatially varying blur detector was applied to extract focused regions from each frame, independent of camera settings. Focus area data was generated every 15 frames and linearly interpolated for intervening frames. Metrics included the position and movement of the focus area relative to the frame center (measured in pixels), the size and variability of the focus area, and the percentage of time surgical tools were present within the focus area

Results: The AI algorithm calculated the average distance between the focus area center and the frame center, starting at approximately 200 pixels and decreasing as the surgery progressed. The standard deviation of the focus area remained stable throughout the operation, indicating consistent movement. Tools were outside the field of focus for less than 15% of the procedure. The focus area size, while small relative to the total frame, remained constant for the duration of the surgery.

Conclusion : We developed an AI-driven field of focus metric to quantify the duration that surgical targets remained in focus. The AI successfully tracked focus areas and tool presence with accuracy, even when interpolating across frames. These metrics, including focus area size and tool time within the focus area, could offer objective insights into the skill progression of surgical trainees. Future work will explore how these metrics correlate with surgeon experience and clinical outcomes, and investigate how these metrics can be used as performance feedback for neurosurgical trainees.