Scholarlink

Background <p>Laparoscopic cholecystectomy (LC) is a fundamental procedure in surgical training. Conventional surgical competency assessment relies heavily on outcome-oriented metrics; however, process-oriented metrics, such as temporal patterns of surgical actions, may better capture operative efficiency and technical proficiency. Nevertheless, existing expert-based evaluation tools remain resource-intensive and susceptible to observer bias.</p> Methods <p>We used the publicly available Cholec80 dataset. Surgical action segments were sampled from dissection and completion stages during Calot’s triangle dissection phase using a fixed-length sliding window. A lag-aligned distance method enabled stage-specific comparison of segments across videos. Representative segments from the high-performance group were used to compute cumulative temporal dissimilarity scores for test videos. These scores were first correlated with ground-truth competency scores to assess overall algorithm performance. SHapley Additive exPlanations (SHAP) analysis identified predictive features from both surgical outcome- and process-oriented metrics, which were subsequently selected to train a support vector regression (SVR) model. Finally, the model’s performance in predicting competency scores was evaluated.</p> Results <p>Eighty LC videos (median competency score: 26; IQR 23.9–28.0) were used for algorithm and model evaluations, with six high-performance videos (scores ≥ 31) selected to establish representative segments. Cumulative temporal dissimilarity scores showed a significant negative correlation with ground-truth scores (<i>r</i> =  − 0.61; <i>P</i> &lt; .001). Critical-view-of-safety score, operative time, and cumulative temporal dissimilarity score were identified as the top three predictors for SVR modeling. The model demonstrated strong predictive performance, with predicted scores highly correlated with ground-truth scores (<i>r</i> = 0.86; <i>P</i> &lt; .001).</p> Conclusion <p>This study introduces a novel process-oriented algorithm for video-based temporal action analysis using a lag-aligned distance method. Furthermore, by integrating outcome- and process-oriented metrics, the proposed model enables accurate and quantitative prediction of surgical competency scores. This approach advances beyond conventional tools, offering a scalable and interpretable solution for automated, objective evaluation of surgical performance.</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

Surgical video-based temporal action analysis algorithm and competency assessment in laparoscopic cholecystectomy: development and exploratory evaluation

  • 期刊论文

  • 发表日期：

    2025年10月21日

查看全文
  • Hung-Hsuan Yen,
  • Ming-Chih Ho,
  • Meng-Han Yang,
  • Yi-Hsiang Hsiao,
  • Hsiang-Wei Huang,
  • Jia-Yuan Huang,
  • Chun-Chieh Huang,
  • Jakey Blue

摘要

Background

Laparoscopic cholecystectomy (LC) is a fundamental procedure in surgical training. Conventional surgical competency assessment relies heavily on outcome-oriented metrics; however, process-oriented metrics, such as temporal patterns of surgical actions, may better capture operative efficiency and technical proficiency. Nevertheless, existing expert-based evaluation tools remain resource-intensive and susceptible to observer bias.

Methods

We used the publicly available Cholec80 dataset. Surgical action segments were sampled from dissection and completion stages during Calot’s triangle dissection phase using a fixed-length sliding window. A lag-aligned distance method enabled stage-specific comparison of segments across videos. Representative segments from the high-performance group were used to compute cumulative temporal dissimilarity scores for test videos. These scores were first correlated with ground-truth competency scores to assess overall algorithm performance. SHapley Additive exPlanations (SHAP) analysis identified predictive features from both surgical outcome- and process-oriented metrics, which were subsequently selected to train a support vector regression (SVR) model. Finally, the model’s performance in predicting competency scores was evaluated.

Results

Eighty LC videos (median competency score: 26; IQR 23.9–28.0) were used for algorithm and model evaluations, with six high-performance videos (scores ≥ 31) selected to establish representative segments. Cumulative temporal dissimilarity scores showed a significant negative correlation with ground-truth scores (r =  − 0.61; P < .001). Critical-view-of-safety score, operative time, and cumulative temporal dissimilarity score were identified as the top three predictors for SVR modeling. The model demonstrated strong predictive performance, with predicted scores highly correlated with ground-truth scores (r = 0.86; P < .001).

Conclusion

This study introduces a novel process-oriented algorithm for video-based temporal action analysis using a lag-aligned distance method. Furthermore, by integrating outcome- and process-oriented metrics, the proposed model enables accurate and quantitative prediction of surgical competency scores. This approach advances beyond conventional tools, offering a scalable and interpretable solution for automated, objective evaluation of surgical performance.