Digital twin-enabled attention-fused deep learning model for cardiovascular disease classification using MRI

Cardiovascular diseases (CVDs) is one of the most debilitating diseases, causing mortality worldwide, particularly in the elderly. Early and accurate diagnosis of CVD types is crucial for risk reduction and effective intervention. In particular, Magnetic Resonance Imaging (MRI) has emerged as the essential diagnostic technique that visualizes the morphology, dynamics, blood flow, and tissue characteristics of the heart. However, the existing methods are often limited in capturing the most relevant features, fails to handle complex and diverse data, resulting in subpar performance. Consequently, this research presents the Hybrid Distributed Attention fused EfficientNet-based Light Gradient Boosting Machine (HDA-ENetLGM) to improve the CVD classification using MRI images. Specifically, the proposed system uses IoT-enabled cloud-based system designed for safe data management and smart analysis of cardiac MRI information. Significantly, the proposed model utilizes the Hybrid Distributed Attention (HDA) in conjunction with hybrid architecture combining EfficientNetB7 and LightGBM to refine the feature representations, contributing to high classification accuracy. Furthermore, the incorporation of LightGBM for decision making provides computationally efficient multi-class classification with better generalization. More specifically, the Digital Twin provides virtual representations of patients and offers improved decision-making and increases the reliability of the model. Ultimately, the BioMistral-7B based Large Language Model (LLM) assists the medical practitioners in smart clinical decision support with providing summaries of the conditions. Experimental results indicates that the proposed HDA-ENetLGM Method achieved 97.82% accuracy, 96.7% specificity, 98.96% sensitivity, 98.07% F1-score, 97.20% precision, and 0.978 MCC for 90% of training with the Sunnybrook Cardiac MRI dataset.