Nathanael Haynes, William Archie, Blake Stephens, Damien Huckleberry, Lon Eskind, David Levi, Jose R Soto, Maria Baimas-George, Dionisios Vrochides
{"title":"Proprietary platform artificial intelligence-based segmentation of surgical videos facilitates review and rating of operative recordings.","authors":"Nathanael Haynes, William Archie, Blake Stephens, Damien Huckleberry, Lon Eskind, David Levi, Jose R Soto, Maria Baimas-George, Dionisios Vrochides","doi":"10.1007/s00464-026-13321-3","DOIUrl":null,"url":null,"abstract":"<p><strong>Background: </strong>Implementation of artificial intelligence (AI) in surgery may revolutionize assessment of operative techniques through video capture, auto-segmentation and auto-analysis. The aim of this study was to utilize recordings of a highly standardized surgical intervention, testing the ability of commercially available AI software to auto-segment operative footage in a clinically meaningful way.</p><p><strong>Methods: </strong>Twenty-five critical steps in liver graft back-table preparations for connection to a commercially available normothermic perfusion liver pump were determined. Five operative videos were captured manually edited. In addition, using the AI-based GoPro auto-edit feature, the last two recordings were auto-segmented for separate analysis and comparison against the manually edited footage. Each manual and auto-segmented video was assessed for: a) inclusion of each one of the 25 steps, (b) technical performance and completion. The assessment was conducted by an independent experienced transplant surgeon using a modified Likert scale (0-not captured, 1-poor, 2-average, 3-excellent).</p><p><strong>Results: </strong>Reviewing and rating AI auto-segmented videos was 16 times faster compared to raw video review. Reviewing and rating AI auto-segmented videos was 96 times faster compared to manually segmented recordings, accounting for time needed to edit raw videos. The average number of steps captured between both raw videos and the manually edited videos was 22.5. The average number of steps captured across the AI auto-segmented videos was 19.5 (87%). The average technical grade for the raw and the manually edited videos was 2.58 ± 0.71, and 2.14 ± 0.99 for the AI auto-segmented videos, denoting a rating congruency of 82.9%.</p><p><strong>Conclusion: </strong>AI-based auto-segmentation is an attractive tool for assessment of surgical performance, as it significantly expedites the editing process to prepare footage for review by subject matter experts. To which extent this approach will be applicable outside highly standardized surgical interventions must be further studied.</p>","PeriodicalId":22174,"journal":{"name":"Surgical Endoscopy And Other Interventional Techniques","volume":" ","pages":""},"PeriodicalIF":3.3000,"publicationDate":"2026-10-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"Surgical Endoscopy And Other Interventional Techniques","FirstCategoryId":"3","ListUrlMain":"https://doi.org/10.1007/s00464-026-13321-3","RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"Q2","JCRName":"SURGERY","Score":null,"Total":0}
引用次数: 0
Abstract
Background: Implementation of artificial intelligence (AI) in surgery may revolutionize assessment of operative techniques through video capture, auto-segmentation and auto-analysis. The aim of this study was to utilize recordings of a highly standardized surgical intervention, testing the ability of commercially available AI software to auto-segment operative footage in a clinically meaningful way.
Methods: Twenty-five critical steps in liver graft back-table preparations for connection to a commercially available normothermic perfusion liver pump were determined. Five operative videos were captured manually edited. In addition, using the AI-based GoPro auto-edit feature, the last two recordings were auto-segmented for separate analysis and comparison against the manually edited footage. Each manual and auto-segmented video was assessed for: a) inclusion of each one of the 25 steps, (b) technical performance and completion. The assessment was conducted by an independent experienced transplant surgeon using a modified Likert scale (0-not captured, 1-poor, 2-average, 3-excellent).
Results: Reviewing and rating AI auto-segmented videos was 16 times faster compared to raw video review. Reviewing and rating AI auto-segmented videos was 96 times faster compared to manually segmented recordings, accounting for time needed to edit raw videos. The average number of steps captured between both raw videos and the manually edited videos was 22.5. The average number of steps captured across the AI auto-segmented videos was 19.5 (87%). The average technical grade for the raw and the manually edited videos was 2.58 ± 0.71, and 2.14 ± 0.99 for the AI auto-segmented videos, denoting a rating congruency of 82.9%.
Conclusion: AI-based auto-segmentation is an attractive tool for assessment of surgical performance, as it significantly expedites the editing process to prepare footage for review by subject matter experts. To which extent this approach will be applicable outside highly standardized surgical interventions must be further studied.
期刊介绍：
Uniquely positioned at the interface between various medical and surgical disciplines, Surgical Endoscopy serves as a focal point for the international surgical community to exchange information on practice, theory, and research.
Topics covered in the journal include:
-Surgical aspects of:
Interventional endoscopy,
Ultrasound,
Other techniques in the fields of gastroenterology, obstetrics, gynecology, and urology,
-Gastroenterologic surgery
-Thoracic surgery
-Traumatic surgery
-Orthopedic surgery
-Pediatric surgery
求助内容：
应助结果提醒方式：
京公网安备 11010802042870号
