Scholarlink

Purpose <p>Laparoscopic cholecystectomy (LC) carries a risk of bile duct injury, mitigated by achieving a critical view of safety. GoNoGoNet, an open-source artificial intelligence model, aims to support surgical training by identifying safe (“Go”) and unsafe (“NoGo”) dissection zones in real time. We sought to evaluate the feasibility of using GoNoGoNet overlays to annotate unsafe dissection behavior during LC. We hypothesized that GoNoGoNet overlays would enable reliable annotation of unsafe dissection behavior during LC, with higher surgeon technical skill associated with fewer NoGo zone intrusions.</p> Methods <p>LC videos were collected under a quality improvement protocol at a safety-net hospital. The hepatocystic triangle dissection phase was isolated and manually annotated for subsequent GoNoGoNet overlay application. The primary outcome was any surgical tool invasion into the NoGo zone. Invasion events were assessed as raw counts, providing a direct and consistent measure of discrete safety–critical errors. Two trained graders assessed all videos using the Global Operative Assessment of Laparoscopic Skills (GOALS), tool invasion counts, and the Parkland Grading Scale (PGS) to adjust for case complexity. Interrater reliability was measured using intraclass correlation coefficients. Spearman’s rank correlation was used to examine the relationship between GOALS scores and NoGo intrusions. A multivariable linear regression model predicted NoGo intrusions based on GOALS scores, controlling for complexity.</p> Results <p>Out of 59 videos, 40 videos were classified as low-complexity (PGS 1–2), and 19 as high-complexity (PGS 3–5). Strong interrater agreement was observed for GOALS (ICC = 0.94), tool invasion (ICC = 0.82), and PGS grades (ICC = 0.89). GOALS scores and NoGo intrusions were significantly negatively correlated (R = − 0.59;&#xa0;<i>p</i> &lt; 0.05), suggesting higher technical skill was associated with fewer unsafe dissection events. This finding was consistent across low- (R = − 0.63; <i>p</i> &lt; 0.05) and high-complexity cases (R = − 0.54;&#xa0;<i>p</i> &lt; 0.05). GOALS score was a significant negative predictor of NoGo intrusions in multivariable regression (β = − 1.91; 95% CI [− 2.69, − 1.13];&#xa0;<i>p</i> &lt; 0.001; R<sup>2</sup> = 0.305), while case complexity was not significant (<i>p</i> = 0.754).</p> Conclusion <p>GoNoGoNet shows promise as a training tool for enhancing LC safety. The use of its overlays to annotate unsafe dissections revealed an inverse correlation with surgical skill level, regardless of case complexity. Further studies are warranted to validate its broader application in minimally invasive surgical education.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Navigating the danger zone: GoNoGoNet can be used to enhance safety and skill in laparoscopic cholecystectomy

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年2月9日

查看全文
  • Bryanna Stukes,
  • Sharanya Vunnava,
  • Sofia Garces Palacios,
  • Shekhar Madhav Khairnar,
  • Kaustubh Gopal,
  • Sydney Vincenti,
  • Alexander Colonna,
  • Ryan P. Dumas,
  • Ganesh Sankaranarayanan

摘要

Purpose

Laparoscopic cholecystectomy (LC) carries a risk of bile duct injury, mitigated by achieving a critical view of safety. GoNoGoNet, an open-source artificial intelligence model, aims to support surgical training by identifying safe (“Go”) and unsafe (“NoGo”) dissection zones in real time. We sought to evaluate the feasibility of using GoNoGoNet overlays to annotate unsafe dissection behavior during LC. We hypothesized that GoNoGoNet overlays would enable reliable annotation of unsafe dissection behavior during LC, with higher surgeon technical skill associated with fewer NoGo zone intrusions.

Methods

LC videos were collected under a quality improvement protocol at a safety-net hospital. The hepatocystic triangle dissection phase was isolated and manually annotated for subsequent GoNoGoNet overlay application. The primary outcome was any surgical tool invasion into the NoGo zone. Invasion events were assessed as raw counts, providing a direct and consistent measure of discrete safety–critical errors. Two trained graders assessed all videos using the Global Operative Assessment of Laparoscopic Skills (GOALS), tool invasion counts, and the Parkland Grading Scale (PGS) to adjust for case complexity. Interrater reliability was measured using intraclass correlation coefficients. Spearman’s rank correlation was used to examine the relationship between GOALS scores and NoGo intrusions. A multivariable linear regression model predicted NoGo intrusions based on GOALS scores, controlling for complexity.

Results

Out of 59 videos, 40 videos were classified as low-complexity (PGS 1–2), and 19 as high-complexity (PGS 3–5). Strong interrater agreement was observed for GOALS (ICC = 0.94), tool invasion (ICC = 0.82), and PGS grades (ICC = 0.89). GOALS scores and NoGo intrusions were significantly negatively correlated (R = − 0.59; p < 0.05), suggesting higher technical skill was associated with fewer unsafe dissection events. This finding was consistent across low- (R = − 0.63; p < 0.05) and high-complexity cases (R = − 0.54; p < 0.05). GOALS score was a significant negative predictor of NoGo intrusions in multivariable regression (β = − 1.91; 95% CI [− 2.69, − 1.13]; p < 0.001; R2 = 0.305), while case complexity was not significant (p = 0.754).

Conclusion

GoNoGoNet shows promise as a training tool for enhancing LC safety. The use of its overlays to annotate unsafe dissections revealed an inverse correlation with surgical skill level, regardless of case complexity. Further studies are warranted to validate its broader application in minimally invasive surgical education.