Copyright © 2026 Authors retain the copyright of this article. This article is an open access article distributed under the Creative Commons Attribution License which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

@article{198531,
        author = {Sri Pravallika Seshabhattar and V. Chandana and M. Tejaswini and K. Bharath Reddy},
        title = {Multi-Model AI Health Assistants for Chronic Disease Diagnosis: The MedFusionAI Framework},
        journal = {International Journal of Innovative Research in Technology},
        year = {2026},
        volume = {12},
        number = {11},
        pages = {10266-10275},
        issn = {2349-6002},
        url = {https://ijirt.org/article?manuscript=198531},
        abstract = {Chronic diseases, including cardiovascular conditions, diabetes, and chronic kidney disease, represent a staggering global health burden and remain primary drivers of mortality and escalating healthcare costs. Effective management is predicated on early, precise risk assessment; however, existing artificial intelligence (AI) methodologies are frequently limited by a reliance on single-modality data and a failure to adequately address the "missing data" problem. Furthermore, many models offer suboptimal modeling of complex inter-modality relationships and suffer from a "black-box" lack of interpretability, which hinders clinical trust. This paper introduces MedFusionAI, a novel deep learning framework for comprehensive multi-modal medical data fusion. By integrating structured electronic health records (EHR) via Multi-Layer Perceptrons (MLPs), sequential lab results and wearable sensor data via Long Short-Term Memory (LSTM) networks, medical imaging via Convolutional Neural Networks (CNNs), and unstructured clinical notes via ClinicalBERT, MedFusionAI captures a holistic patient profile. The system employs a powerful hybrid fusion scheme that couples early feature-level concatenation with a cross-modal attention mechanism to explicitly weigh the importance of each modality and map their interdependencies. To overcome real-world data sparsity, our framework incorporates a robust modality dropout strategy during training and KNN interpolation with autoencoder-based weights, ensuring diagnostic stability even in the presence of incomplete patient records. Consequently, MedFusionAI achieves a state-of-the-art classification accuracy of 98.76% and an AUC-ROC of 99.18% across all risk classes. By offering Explainable AI capabilities, MedFusionAI provides transparent, interpretable clinical decision support, identifying risk trajectories with high sensitivity to improve preventive care outcomes and reduce diagnostic delays.},
        keywords = {Multi-Modal Data Fusion, Deep Learning, Chronic Disease Prediction, Attention Mechanism, Clinical Decision Support},
        month = {April},
        }