Explainable Multi-Source Data Fusion Model for Early Prediction of Complex Chronic Diseases

Abstract

Complex chronic diseases, by virtue of their progressive nature, are a major challenge in contemporary healthcare because most of them are difficult to predict at early stages, and are characterized by high burden to patients and health care systems. Although artificial intelligence has developed, current predictive models tend to use a single-source of information, so they can only capture the multifactorial nature of chronic conditions. Moreover, fragmented nature of healthcare data such as electronic health records (EHR), clinical variables and medical imaging present a major challenge to successful integration and analysis. In a bid to overcome these shortcomings, this paper will present a more explainable multi-source deep learning fusion model aimed at incorporating heterogeneous healthcare data to precisely and early detect diseases. The presented framework implements the multimodal architecture that processes different types of data with the help of specific deep learning modules and integrates them into each other with the help of a feature-level fusion strategy. In addition, explainable artificial intelligence (XAI) methods, such as SHAP-based feature attribution and visualization tools, are included to promote transparent models with clinical interpretability. The model has been tested on standard healthcare datasets and compared to the traditional machine learning and single-modal deep learning methods. The experimental findings indicate that the proposed fusion model has better performance, and the accuracy, recall (sensitivity) and area under the ROC curve (AUC) are significantly improved. These findings demonstrate the power of the multimodal integration to describe complex patterns of diseases and enhance the ability to detect them earlier. In general, the framework suggested is a strong and interpretable predictive solution, which has immense capacity to be implemented in clinical decision support systems and to further the evolution of intelligent and data-driven applications in healthcare.