Artificial-Intelligence-based Surgical Phase Recognition in Robot-Assisted Radical Prostatectomy and Cross-Surgeon Validation.

Abstract

Background: Artificial intelligence (AI) has shown potential in various fields; however, its practical application in surgery remains limited. We developed an AI system capable of automatically recognizing surgical phases in robot-assisted radical prostatectomy (RARP) and confirmed its accuracy through cross-surgeon validation.

Materials and methods: We analyzed clinical data from 102 patients who underwent RARP, including 81 consecutive patients operated on by one surgeon (surgeon A) and 21 operated on by five other surgeons (surgeons B-F). In total, 65 of the 81 patients were used for AI development, while the remaining 16, in addition to the 21 patients operated on by surgeons B-F, were used for AI validation. We classified surgical operations into nine phases. Well-trained surgeons annotated the time corresponding to each surgical phase for each video. We used Temporal Convolutional Networks for the Operating Room (TeCNO) to develop the AI model and evaluated its precision.

Results: In AI development, 919,231 frames were utilized. Testing involved 216,357 frames from surgeon A and 249,553 frames from surgeons B-F. When the developed AI was used to analyze surgical videos from surgeon A, precision reached 0.94. In contrast, when the AI was applied to videos from surgeons B-F, precision was 0.83.

Conclusions: The AI we developed not only showed high accuracy, but also demonstrated generalizability across different surgeons. By comprehensively evaluating surgical videos, our AI may be used to assess the quality of surgeries, thereby providing valuable feedback to surgeons and enhancing the effectiveness of surgical education.