Scholarlink

<p>Liver disease (LD) is a major global health concern, causing approximately 2&#xa0;million deaths annually. As a vital organ, the liver regulates essential metabolic processes and overall health, making an early and accurate forecast of liver disease vital for effective intervention. Accurate liver disease prediction remains challenging due to complex nonlinear relationships among features and the limitations of conventional algorithms in efficiently capturing these interactions. To address these challenges, this study proposes an advanced framework to predict liver diseases using Multi-View Heterogeneous Graph Neural Networks (PLD-MHGNN) for early diagnosis. Clinical data from the Indian Liver Patient Dataset were pre-processed using Cauchy Robust Correction-Sage Husa Extended Kalman Filtering (CRCSHEKF) to handle missing values, normalize and standardize the dataset. The Eel and Grouper Optimizer (EGO) was employed for feature selection to identify the most informative clinical markers, capturing complex inter-feature relationships. The selected features were then input to a Multi-view Heterogeneous Graph Neural Network (MHGNN) for accurate prediction of liver disease. The proposed PLD-MHGNN illustrated higher performance, reaching an accuracy of 98.42%, precision of 98.15%, recall of 97.88%, F1-score of 98.12%, specificity of 97.45%, AUC of 0.9875, and an error rate of 1.42%, outperforming existing techniques like a comparative analysis of machine learning (ML) techniques with tree-structured parzen estimator for prediction of liver disease (TSP-RF-LDP), Machine learning methods for forecasting metabolic dysfunction-associated steatotic liver disease prevalence utilizing basic demographic as well as clinical characteristics (PMD-LDP-XGBoost) and Prediction of chronic liver disease patients utilizing integrated projection based statistical feature extraction with ML techniques (PCLDP-CMVO-IPSFE) respectively. This framework provides a reliable and robust tool for early liver disease detection, with potential clinical applicability for improved patient management and intervention planning.</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

Advanced prediction and diagnosis of liver diseases using multi-view heterogeneous graph neural networks

  • 期刊论文

  • 发表日期：

    2025年11月20日

查看全文
  • Yashwant Dongre,
  • Priya M Shelke,
  • Suruchi Dedgaonkar,
  • Geeta S. Navale,
  • Amol V. Dhumane

摘要

Liver disease (LD) is a major global health concern, causing approximately 2 million deaths annually. As a vital organ, the liver regulates essential metabolic processes and overall health, making an early and accurate forecast of liver disease vital for effective intervention. Accurate liver disease prediction remains challenging due to complex nonlinear relationships among features and the limitations of conventional algorithms in efficiently capturing these interactions. To address these challenges, this study proposes an advanced framework to predict liver diseases using Multi-View Heterogeneous Graph Neural Networks (PLD-MHGNN) for early diagnosis. Clinical data from the Indian Liver Patient Dataset were pre-processed using Cauchy Robust Correction-Sage Husa Extended Kalman Filtering (CRCSHEKF) to handle missing values, normalize and standardize the dataset. The Eel and Grouper Optimizer (EGO) was employed for feature selection to identify the most informative clinical markers, capturing complex inter-feature relationships. The selected features were then input to a Multi-view Heterogeneous Graph Neural Network (MHGNN) for accurate prediction of liver disease. The proposed PLD-MHGNN illustrated higher performance, reaching an accuracy of 98.42%, precision of 98.15%, recall of 97.88%, F1-score of 98.12%, specificity of 97.45%, AUC of 0.9875, and an error rate of 1.42%, outperforming existing techniques like a comparative analysis of machine learning (ML) techniques with tree-structured parzen estimator for prediction of liver disease (TSP-RF-LDP), Machine learning methods for forecasting metabolic dysfunction-associated steatotic liver disease prevalence utilizing basic demographic as well as clinical characteristics (PMD-LDP-XGBoost) and Prediction of chronic liver disease patients utilizing integrated projection based statistical feature extraction with ML techniques (PCLDP-CMVO-IPSFE) respectively. This framework provides a reliable and robust tool for early liver disease detection, with potential clinical applicability for improved patient management and intervention planning.