A Novel Deep Neural Network Model for Multi-Label Chronic Disease Prediction

Area of Science:

• Medical Informatics
• Artificial Intelligence
• Computational Biology

Background:

• Chronic diseases pose a significant threat to human health.
• Early prediction and intervention are critical for managing chronic diseases.
• Accurate multi-label classification of chronic diseases remains a challenge.

Purpose of the Study:

• To develop a novel deep learning model for predicting multiple chronic diseases simultaneously.
• To investigate the effectiveness of problem transformation methods for multi-label chronic disease prediction.
• To introduce a new loss function that accounts for correlations between diseases.

Main Methods:

• The study transformed chronic disease prediction into a multi-label classification problem.
• A novel convolutional neural network (CNN) architecture, GroupNet, was proposed.
• Binary Relevance (BR) and Label Powerset (LP) methods were used for label transformation.
• A correlated loss function, integrating disease correlation coefficients, was developed for GroupNet.

Main Results:

• GroupNet achieved a top accuracy of 81.13% in chronic disease prediction.
• The proposed GroupNet model outperformed existing methods in multi-label classification tasks.
• Experiments were conducted on real-world physical examination datasets.

Conclusions:

• The GroupNet model demonstrates superior performance in multi-label chronic disease classification.
• Integrating disease correlations into the loss function enhances prediction accuracy.
• The proposed approach offers a promising tool for early detection and management of chronic diseases.