Scholarlink

Background <p>Microvascular invasion (MVI) is a crucial risk factor for postoperative recurrence in patients with hepatocellular carcinoma (HCC). Accurate preoperative assessment of MVI and appropriate resection margins (RM) could reduce recurrence rates of HCC.</p> Methods <p>A total of 1651 eligible patients who underwent liver resection between January 1, 2018 and June 30, 2023 were retrospectively collected. They were divided into the training (830 patients), the validation (356 patients), and the separate internal temporal test (465 patients) sets. The readily available features are emphasized in this study (including basic clinical and hematological features). Lasso regression was used for features' selection. Multivariate logistic regression and restricted cubic splines were conducted to evaluate the associations between the selected features and MVI. Seven machine learning algorithms were employed to constructed predictive model.</p> Results <p>The preoperative factors associated with MVI were age, thrombin time, international normalized ratio (INR), aspartate aminotransferase, direct bilirubin, alpha-fetoprotein, mean corpuscular volume, multi tumor, tumor size, and fibrinogen/INR. Incorporating these factors, the LightGBM model achieved good predictive efficacy than other models, with the AUCs of 0.84 (95% CI: 0.81, 0.87) and 0.78 (95% CI: 0.73, 0.83) in the training and validation sets. This model also maintained satisfactory prediction in the separate temporal test set (AUC: 0.78, 95% CI: 0.74, 0.83) with good calibration and clinical net benefit. In addition, the recurrence-free survival (RFS) could be significantly stratified in the MVI high-risk and low-risk groups by this model, and the RFS of patients with wide RM was higher than that of patients with narrow RM in the high-risk MVI group (&lt; 1&#xa0;cm).</p> Conclusion <p>The LightGBM model using readily available features was constructed to rapidly assess the risk of MVI in HCC patients. Based on this model, we developed a web-independent calculator for clinicians to easily operate the proposed model.</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

An easily machine learning-based tool for preliminary risk assessment of microvascular invasion in hepatocellular carcinoma

  • 期刊论文

  • 发表日期：

    2025年8月27日

查看全文
  • Shaoming Song,
  • Feifan Wu,
  • Gong Zhang,
  • Haoming Xu,
  • Guineng Zeng,
  • Lin Zhu,
  • Ruiqiu Chen,
  • Zizheng Wang,
  • Xiuping Zhang,
  • Rong Liu

摘要

Background

Microvascular invasion (MVI) is a crucial risk factor for postoperative recurrence in patients with hepatocellular carcinoma (HCC). Accurate preoperative assessment of MVI and appropriate resection margins (RM) could reduce recurrence rates of HCC.

Methods

A total of 1651 eligible patients who underwent liver resection between January 1, 2018 and June 30, 2023 were retrospectively collected. They were divided into the training (830 patients), the validation (356 patients), and the separate internal temporal test (465 patients) sets. The readily available features are emphasized in this study (including basic clinical and hematological features). Lasso regression was used for features' selection. Multivariate logistic regression and restricted cubic splines were conducted to evaluate the associations between the selected features and MVI. Seven machine learning algorithms were employed to constructed predictive model.

Results

The preoperative factors associated with MVI were age, thrombin time, international normalized ratio (INR), aspartate aminotransferase, direct bilirubin, alpha-fetoprotein, mean corpuscular volume, multi tumor, tumor size, and fibrinogen/INR. Incorporating these factors, the LightGBM model achieved good predictive efficacy than other models, with the AUCs of 0.84 (95% CI: 0.81, 0.87) and 0.78 (95% CI: 0.73, 0.83) in the training and validation sets. This model also maintained satisfactory prediction in the separate temporal test set (AUC: 0.78, 95% CI: 0.74, 0.83) with good calibration and clinical net benefit. In addition, the recurrence-free survival (RFS) could be significantly stratified in the MVI high-risk and low-risk groups by this model, and the RFS of patients with wide RM was higher than that of patients with narrow RM in the high-risk MVI group (< 1 cm).

Conclusion

The LightGBM model using readily available features was constructed to rapidly assess the risk of MVI in HCC patients. Based on this model, we developed a web-independent calculator for clinicians to easily operate the proposed model.