文献速递胃肠外科围术期管理与 ERAS

生物物理建模与人工智能用于腹腔镜低位直肠前切除吻合口血供定量评估

Biophysical modeling and artificial intelligence for quantitative assessment of anastomotic blood supply in laparoscopic low anterior rectal resection.
PubMed 文献 · Surg Endosc · 2025-04-14

本研究对68例低位直肠癌患者的腹腔镜荧光视频进行分析,沿结肠带选取1263个测量点,结合荧光曲线分析、生物物理建模与机器学习自动评分吻合口灌注质量。XGBoost模型测试集RMSE为2.47、MAE为1.99、R2为97.2%,远端吻合口低灌注评分与术后并发症显著相关(P < 0.001)。

要点

  • 生物物理建模可定量灌注动态,超越传统时间因子分析
  • 机器学习自动评分达到术者级准确度,分析可在2分钟内完成
  • 远端吻合口低灌注评分与术后并发症显著相关

核心结论

  • 该框架通过生物物理建模、机器学习与自动化流程提升结直肠手术术中灌注评估与决策。

研究设计与样本

  • 分析68例低位直肠癌患者腹腔镜荧光视频,每例选取15-20个测量点,共1263个测量点。

主要发现

  • XGBoost模型测试集RMSE为2.47、MAE为1.99、R2为97.2%,优于传统时间因子分析。
  • 模型在年龄、糖尿病或吸烟亚组间预测差异无统计学意义(P > 0.05)。
  • 远端吻合口区域低灌注评分与术后并发症显著相关(P < 0.001)。
  • 自动化软件在2分钟内完成分析。

关键数值与推荐

  • 三名临床医生独立使用形态学特征与生物物理指标对灌注质量进行0-100评分。
  • 采用RANSAC算法与非线性最小二乘拟合提取生物物理参数。

临床怎么用

  • 可在术中对吻合口灌注进行快速定量评估,辅助判断吻合口质量与决策。

局限性

  • 样本量68例,单中心研究,外部验证尚未报告。

以上解读依据论文摘要(全文未开放获取或许可不允许翻译转载)。

原文

PURPOSE: Fluorescence imaging is critical for intraoperative intestinal perfusion assessment in colorectal surgery, yet its clinical adoption remains limited by subjective interpretation and lack of quantitative standards. This study introduces an integrated approach combining fluorescence curve analysis, biophysical modeling, and machine learning to improve intraoperative perfusion assessment. METHODS: Laparoscopic fluorescence videos from 68 low rectal cancer patients were analyzed, with 1,263 measurement points (15-20 per case) selected along colonic bands. Fluorescence intensity dynamics were extracted via color space transformation, video stabilization and image registration, then modeled using the Random Sample Consensus (RANSAC) algorithm and nonlinear least squares fitting to derive biophysical parameters. Three clinicians independently scored perfusion quality (0-100 scale) using morphological features and biophysical metrics. An XGBoost model was trained on these parameters to automate scoring. RESULTS: The model achieved superior test performance, with a root mean square error (RMSE) of 2.47, a mean absolute error (MAE) of 1.99, and an R2 of 97.2%, outperforming conventional time-factor analyses. It demonstrated robust generalizability, showing no statistically significant prediction differences across age, diabetes, or smoking subgroups (P > 0.05). Clinically, low perfusion scores in distal anastomotic regions were significantly associated with postoperative complications (P < 0.001), validating the scoring system's clinical relevance and discriminative capacity. The automated software we developed completed analyses within 2 min, enabling rapid intraoperative assessment. CONCLUSION: This framework synergistically enhances surgical evaluation through three innovations: (1) Biophysical modeling quantifies perfusion dynamics beyond time-based parameters; (2) Machine learning integrates multimodal data for surgeon-level accuracy; (3) Automated workflow enables practical clinical translation. By addressing limitations of visual assessment through quantitative, rapid, and generalizable analysis, this method advances intraoperative perfusion monitoring and decision-making in colorectal surgery.

来源:Surg Endosc,查看原文。版权归原作者所有。

引用

  1. He W, Zhu H, Rao X, et al. Biophysical modeling and artificial intelligence for quantitative assessment of anastomotic blood supply in laparoscopic low anterior rectal resection. Surg Endosc. 2025. PMID: 40227485. doi:10.1007/s00464-025-11693-6

供执业医师工作参考,不替代临床判断;数值与分级以指南原文为准。