Scholarlink

<p>Despite significant advancements in the diagnosis and treatment of Hepatocellular carcinoma (HCC), the overall prognosis remains challenging due to its multifactorial nature. Recently, machine learning has been increasingly utilized in HCC-related research, offering significant advancements in diagnostic accuracy and early detection. However, the application of these techniques in diverse populations, particularly underrepresented groups, remains limited. In this study, seven machine learning algorithms: logistic regression (LR), k-nearest neighbor (KNN), support vector classifier (SVC), decision tree (DT), random forest (RF), stochastic gradient boosting (GB) and LightGBM, were tested. Optimal hyperparameters were determined using a grid-search approach, and several feature selection techniques were applied to identify key predictors in detecting HCC. Among the tested algorithms, the RF, with its optimal hyperparameters, demonstrated superior predictive performance at 98.9 % accuracy, 90.5% sensitivity, 99.8% specificity, 99% NPV, 98% PPV, and an AUC of 0.99. LightGBM delivered similarly strong results, attaining an AUC of 0.99, 99.1% accuracy, 94.9% sensitivity, 99.5% specificity, 99.5% NPV, and 95.5% PPV. Notably, only seven clinical predictors: age, albumin, alkaline phosphatase (ALP), alpha-fetoprotein (AFP), des-gamma-carboxy prothrombin (DCP), aspartate transaminase, and platelet count, were necessary to achieve these results. In addition, we illustrate that RF has a robust capability to detect HCC irrespective of the clinical risk factors. This study underscores the potential of machine-learning-based detection models as an alternative for HCC detection and screening. The use of a limited set of predictors not only simplifies the diagnostic process but is also advantageous for resource-limited and remote settings. Using machine learning in HCC detection can significantly improve diagnostic efficiency and enable earlier detection, potentially leading to better patient outcomes and more effective treatment strategies. This research is the first comprehensive analysis of HCC clinical data from a Filipino cohort, offering critical insights into the disease within an underrepresented group.</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

Minimal clinical predictors enable machine learning detection of hepatocellular carcinoma in a Filipino cohort

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年11月11日

查看全文
  • Ranzivelle Marianne L. Roxas-Villanueva,
  • Mojhune Gabriel Manzanillo,
  • Edhel S. Tripon,
  • Dannah Zemirah G. Junio,
  • Marc Daniel O. Jamindang,
  • Ramonito B. Nuique,
  • Beatrice J. Tiangco,
  • Jason R. Albia,
  • Christian M. Alis

摘要

Despite significant advancements in the diagnosis and treatment of Hepatocellular carcinoma (HCC), the overall prognosis remains challenging due to its multifactorial nature. Recently, machine learning has been increasingly utilized in HCC-related research, offering significant advancements in diagnostic accuracy and early detection. However, the application of these techniques in diverse populations, particularly underrepresented groups, remains limited. In this study, seven machine learning algorithms: logistic regression (LR), k-nearest neighbor (KNN), support vector classifier (SVC), decision tree (DT), random forest (RF), stochastic gradient boosting (GB) and LightGBM, were tested. Optimal hyperparameters were determined using a grid-search approach, and several feature selection techniques were applied to identify key predictors in detecting HCC. Among the tested algorithms, the RF, with its optimal hyperparameters, demonstrated superior predictive performance at 98.9 % accuracy, 90.5% sensitivity, 99.8% specificity, 99% NPV, 98% PPV, and an AUC of 0.99. LightGBM delivered similarly strong results, attaining an AUC of 0.99, 99.1% accuracy, 94.9% sensitivity, 99.5% specificity, 99.5% NPV, and 95.5% PPV. Notably, only seven clinical predictors: age, albumin, alkaline phosphatase (ALP), alpha-fetoprotein (AFP), des-gamma-carboxy prothrombin (DCP), aspartate transaminase, and platelet count, were necessary to achieve these results. In addition, we illustrate that RF has a robust capability to detect HCC irrespective of the clinical risk factors. This study underscores the potential of machine-learning-based detection models as an alternative for HCC detection and screening. The use of a limited set of predictors not only simplifies the diagnostic process but is also advantageous for resource-limited and remote settings. Using machine learning in HCC detection can significantly improve diagnostic efficiency and enable earlier detection, potentially leading to better patient outcomes and more effective treatment strategies. This research is the first comprehensive analysis of HCC clinical data from a Filipino cohort, offering critical insights into the disease within an underrepresented group.