Surgical Video Platform: An Online Artificial Intelligence (AI) Surgical Video Database Used in the Analysis of Endoscopic Endonasal Transsphenoidal Pituitary Surgery

Introduction: The endoscopic endonasal transsphenoidal technique has been adopted as the gold standard, minimally invasive technique in pituitary surgery. Use of surgical videos as supplementary to teaching and combined with AI analysis can optimize a thorough understanding of the anatomy, approach, closure, and other components of this type of surgery.

Methods: A prospectively maintained surgical video library was used to choose 66 endoscopic endonasal transsphenoidal pituitary surgical cases from our institution and was reviewed based on tumor type, approach, and excluding reoperations. The AI was then trained using these videos to generate a tool heat map, graphs based on tool appearance, and procedure breakdown including a nasal, sphenoid, sellar, and closure phase.

Results: Of 122 surgical videos available on the database from our institution and 96 pituitary surgeries available, 66 videos met the inclusion criteria for videos that would be chosen to train the AI model. The average time for all pituitary cases was 84 minutes with all cases having a nasal, sphenoid, sellar, and closure phase. All heat maps, tool appearance graphs, and other figures generated for the chosen procedures demonstrated consistent tool usage outcomes with the suction being the dominant tool used.

Conclusion : The integration of AI and surgical videos into the surgical teaching environment may optimize the creation of an understanding of the surgical setting and minimize mistakes during teaching.