A.T.L.A.S (Applied Technology Laboratory for Advanced Surgery) Program, Department of Urology, Roswell Park Comprehensive Cancer Center, Elm & Carlton St, Buffalo, NY, 14263, USA.
Department of Industrial and Systems Engineering, University at Buffalo, Buffalo, NY, USA.
Int J Comput Assist Radiol Surg. 2019 Apr;14(4):697-707. doi: 10.1007/s11548-018-1881-9. Epub 2018 Nov 20.
To develop and validate an automated assessment of surgical performance (AASP) system for objective and computerized assessment of pelvic lymph node dissection (PLND) as an integral part of robot-assisted radical cystectomy (RARC) using console-feed videos recorded during live surgery.
Video recordings of 20 PLNDs were included. The quality of lymph node clearance was assessed based on the features derived from the computer vision process which include: the number and cleared area of the vessels/nerve (N-Vs); image median color map; and mean entropy (measures the level of disorganization) in the video frame. The automated scores were compared to the validated pelvic lymphadenectomy appropriateness and completion evaluation (PLACE) scoring rated by a panel of expert surgeons. Logistic regression analysis was employed to compare automated scores versus PLACE scores.
Fourteen procedures were used to develop the AASP algorithm. A logistic regression model was trained and validated using the aforementioned features with 30% holdout cross-validation. The model was tested on the remaining six procedures, and the accuracy of predicting the expert-based PLACE scores was 83.3%.
To our knowledge, this is the first automated surgical skill assessment tool that provides an objective evaluation of surgical performance with high accuracy compared to expert surgeons' assessment that can be extended to any endoscopic or robotic video-enabled surgical procedure.
开发并验证一种自动化手术绩效评估（AASP）系统，用于对机器人辅助根治性膀胱切除术（RARC）中作为整体部分的盆腔淋巴结清扫术（PLND）进行客观和计算机化评估，该系统使用手术过程中记录的控制台馈送视频。
纳入 20 例 PLND 视频记录。根据计算机视觉过程中得出的特征评估淋巴结清除质量，这些特征包括：血管/神经的数量和清除区域（N-Vs）；图像中值颜色图；以及视频帧中的平均熵（衡量无序程度）。自动评分与由专家外科医生组成的小组进行的验证后的盆腔淋巴结切除术适当性和完成评估（PLACE）评分进行比较。采用逻辑回归分析比较自动评分与 PLACE 评分。
使用上述特征开发了 AASP 算法，其中 14 个程序用于开发算法。使用上述特征并结合 30%的保留交叉验证对逻辑回归模型进行了训练和验证。该模型在其余 6 个程序上进行了测试，预测专家基于 PLACE 评分的准确率为 83.3%。
据我们所知，这是第一个提供与专家评估相比具有高精度的客观手术绩效评估的自动化手术技能评估工具，并且可以扩展到任何内镜或机器人视频支持的手术程序。
