Scholarlink

Background <p>Laparoscopic cholecystectomy outcomes are significantly influenced by gallbladder inflammation severity, yet current intraoperative assessment remains subjective and lacks standardization. This study aimed to develop and validate an AI framework (ENDOLAP-IA) for real-time, objective severity classification to predict procedural difficulty and complications.</p> Methods <p>A prospective cohort study (July 2023–February 2024) enrolled 53 elective cholecystectomy patients. A standardized 9-item checklist (ENDOLAP-IA tool) was validated for image quality via Delphi consensus and pilot testing. Intraoperative images (<i>n</i> =  &gt; 2000) were captured at three procedural stages using a 1080p laparoscope and annotated per Parkland severity grades. A YOLOv8 model was fine-tuned using transfer learning. Validation included fivefold cross-validation and external testing with 23 images assessed by blinded surgeons. Performance metrics (accuracy, precision, recall, F1-score, AUC-ROC) were evaluated against expert consensus.</p> Results <p>The ENDOLAP tool achieved excellent content validity (CVI &gt; 0.85) and inter-rater reliability (ICC = 0.82). The AI model demonstrated strong overall accuracy (87.3% ± 3.2%) and discriminative capability (AUC = 0.923, 95% CI [0.896–0.945]). Precision and recall were 84.7 and 86.8%, respectively. Performance varied by severity grade, with highest accuracy in Grades 1–2 (92.4–89.7%) and lower but clinically acceptable accuracy in severe Grades 4–5 (79.3–81.6%). External validation showed 82.6% agreement with surgeons and 91.3% sensitivity for severe inflammation. The AI eliminated interobserver variability seen in 17% of human assessments.</p> Conclusions <p>ENDOLAP-IA is the first AI framework to standardize intraoperative inflammatory severity classification in laparoscopic cholecystectomy. It achieves clinically reliable performance, enhances objectivity, and enables real-time decision support. Integration into surgical workflows may improve safety, training, and comparative outcome analyses.</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

Development and validation of the ENDOLAP artificial intelligence framework for inflammation severity classification in laparoscopic cholecystectomy: a cross-sectional study

  • 期刊论文

  • 发表日期：

    2025年8月20日

查看全文
  • Norman A. Rendón Mejía,
  • Said De la Cruz Rey,
  • Carlos R. Cervantes-Sánchez,
  • Eduardo Cañedo Figueroa,
  • Graciela Ramírez Alonso

摘要

Background

Laparoscopic cholecystectomy outcomes are significantly influenced by gallbladder inflammation severity, yet current intraoperative assessment remains subjective and lacks standardization. This study aimed to develop and validate an AI framework (ENDOLAP-IA) for real-time, objective severity classification to predict procedural difficulty and complications.

Methods

A prospective cohort study (July 2023–February 2024) enrolled 53 elective cholecystectomy patients. A standardized 9-item checklist (ENDOLAP-IA tool) was validated for image quality via Delphi consensus and pilot testing. Intraoperative images (n =  > 2000) were captured at three procedural stages using a 1080p laparoscope and annotated per Parkland severity grades. A YOLOv8 model was fine-tuned using transfer learning. Validation included fivefold cross-validation and external testing with 23 images assessed by blinded surgeons. Performance metrics (accuracy, precision, recall, F1-score, AUC-ROC) were evaluated against expert consensus.

Results

The ENDOLAP tool achieved excellent content validity (CVI > 0.85) and inter-rater reliability (ICC = 0.82). The AI model demonstrated strong overall accuracy (87.3% ± 3.2%) and discriminative capability (AUC = 0.923, 95% CI [0.896–0.945]). Precision and recall were 84.7 and 86.8%, respectively. Performance varied by severity grade, with highest accuracy in Grades 1–2 (92.4–89.7%) and lower but clinically acceptable accuracy in severe Grades 4–5 (79.3–81.6%). External validation showed 82.6% agreement with surgeons and 91.3% sensitivity for severe inflammation. The AI eliminated interobserver variability seen in 17% of human assessments.

Conclusions

ENDOLAP-IA is the first AI framework to standardize intraoperative inflammatory severity classification in laparoscopic cholecystectomy. It achieves clinically reliable performance, enhances objectivity, and enables real-time decision support. Integration into surgical workflows may improve safety, training, and comparative outcome analyses.