基于DBN-ELM的肝病诊断模型研究

Research on Liver Disease Diagnosis Model Based on DBN-ELM

摘要: 肝病是一种具有高度传染性和致命性的疾病，传统的肝病诊断方法通常是耗时或具有侵入性的成像技术和活检，而深度学习提供了一种有效且无侵入性的检测途径。本研究基于UCI机器学习库的肝病公开数据集，其包含416个健康样本和167个肝病样本以及10个血液检测临床特征。本研究使用主成分分析(Principal Component Analysis, PCA)提取了重要特征，经过数据预处理后，构建并训练了一个基于深度信念网络–极限学习机(Deep Belief Network-Extreme Learning Machine, DBN-ELM)的肝病检测模型，该模型结合了DBN的无监督预训练和ELM的快速训练能力。DBN-ELM模型在肝病数据集上分别取得了87.31%的准确率、65.27%的敏感度、96.15%的特异性、72.73%的F1-Score和0.9188的受试者工作特征曲线下面积(Area under the Receiver Operating Characteristic Curve, AUC)。这些较好的性能表明，DBN-ELM可以有效地检测肝脏疾病，帮助临床医生实现肝病的早期诊断。

Abstract: Liver disease is a highly contagious and deadly disease. Traditional diagnostic methods for liver disease are usually time-consuming or invasive imaging techniques and biopsies, while deep learning provides an effective and non-invasive detection pathway. This study is based on the liver disease public dataset of UCI machine learning library, which includes 416 healthy samples, 167 liver disease samples, and 10 clinical features of blood tests. This study used Principal Component Analysis (PCA) to extract important features, and after data preprocessing, constructed and trained a liver disease detection model based on Deep Belief Network Extreme Learning Machine (DBN-ELM), which combines the unsupervised pre-training of DBN and the fast training ability of ELM. The DBN-ELM model achieved an accuracy of 87.31%, sensitivity of 65.27%, specificity of 96.15%, F1-Score of 72.73%, and Area under the Receiver Operating Characteristic Curve (AUC) of 0.9188 on the liver disease dataset. These good performances indicate that DBN-ELM can effectively detect liver diseases and help clinical doctors achieve early diagnosis of liver diseases.