Scholarlink

<p>Accurate evaluation of liver fibrosis is essential for surgical planning in patients with hepatocellular carcinoma (HCC), as advanced fibrosis or cirrhosis can significantly alter resection strategy and postoperative outcomes. However, preoperative assessments using blood-based indices or imaging often fail to accurately reflect the true degree of fibrosis observed intraoperatively. This study aimed to develop and evaluate a deep learning model for video-based diagnostic analysis for of severe liver fibrosis (F3–F4), compared to surgeons’ visual assessment and conventional non-invasive scores. In this single-center retrospective study, we included 103 patients who underwent minimally invasive liver resection for HCC between December 2019 and March 2022. Intraoperative video frames were extracted and labeled according to METAVIR fibrosis grades. A DenseNet-121 architecture pretrained on ImageNet was fine-tuned to classify severe (F3–F4) vs. non-severe (F0–F2) fibrosis using five-fold cross-validation. The model’s performance was compared against five experienced liver surgeons’ visual estimations and two commonly used indices: APRI and FIB-4. Diagnostic metrics included area under the receiver operating characteristic curve (AUROC), sensitivity, specificity, and F1-score. The deep learning model achieved a mean (± SD) AUROC of 0.927 (± 0.039), with sensitivity of 0.918 (± 0.110) and specificity of 0.910 (± 0.032). Surgeons exhibited lower AUROCs (0.844 for more experienced surgeons and 0.808 for less experienced), primarily due to lower specificity. APRI and FIB-4 also showed inferior discriminative capabilities, with AUROCs of 0.680 and 0.670, respectively. A deep learning approach using intraoperative liver surface images demonstrated superior performance in detecting severe liver fibrosis compared to surgeons’ assessments and standard non-invasive indices. These findings suggest a potential to provide objective reference data to determine intraoperative surgical strategies for HCC. Prospective multicenter validation is warranted to confirm generalizability and clinical impact.</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

Intraoperative video-based artificial intelligence model exceeding surgeon accuracy for predicting severe fibrosis in minimally invasive liver surgery

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年4月9日

查看全文
  • Namkee Oh,
  • Bogeun Kim,
  • Sunghyo An,
  • Eunjin Lee,
  • Hayeon Do,
  • Jiyoung Baik,
  • Suk Min Gwon,
  • Jinsoo Rhu,
  • Jongman Kim,
  • Inwoo Hwang,
  • Sang Yun Ha,
  • Myung Jin Chung,
  • Hakje Yoo,
  • Gyu-Seong Choi

摘要

Accurate evaluation of liver fibrosis is essential for surgical planning in patients with hepatocellular carcinoma (HCC), as advanced fibrosis or cirrhosis can significantly alter resection strategy and postoperative outcomes. However, preoperative assessments using blood-based indices or imaging often fail to accurately reflect the true degree of fibrosis observed intraoperatively. This study aimed to develop and evaluate a deep learning model for video-based diagnostic analysis for of severe liver fibrosis (F3–F4), compared to surgeons’ visual assessment and conventional non-invasive scores. In this single-center retrospective study, we included 103 patients who underwent minimally invasive liver resection for HCC between December 2019 and March 2022. Intraoperative video frames were extracted and labeled according to METAVIR fibrosis grades. A DenseNet-121 architecture pretrained on ImageNet was fine-tuned to classify severe (F3–F4) vs. non-severe (F0–F2) fibrosis using five-fold cross-validation. The model’s performance was compared against five experienced liver surgeons’ visual estimations and two commonly used indices: APRI and FIB-4. Diagnostic metrics included area under the receiver operating characteristic curve (AUROC), sensitivity, specificity, and F1-score. The deep learning model achieved a mean (± SD) AUROC of 0.927 (± 0.039), with sensitivity of 0.918 (± 0.110) and specificity of 0.910 (± 0.032). Surgeons exhibited lower AUROCs (0.844 for more experienced surgeons and 0.808 for less experienced), primarily due to lower specificity. APRI and FIB-4 also showed inferior discriminative capabilities, with AUROCs of 0.680 and 0.670, respectively. A deep learning approach using intraoperative liver surface images demonstrated superior performance in detecting severe liver fibrosis compared to surgeons’ assessments and standard non-invasive indices. These findings suggest a potential to provide objective reference data to determine intraoperative surgical strategies for HCC. Prospective multicenter validation is warranted to confirm generalizability and clinical impact.