Attention-Guided Multimodal Fusion using Hybrid CNN-LSTM Networks for Enhanced Medical Diagnosis

Accurate and timely diagnosis of medical conditions often requires integrating diverse types of data, such as physiological signals and medical imaging. This paper presents an attention-guided multimodal fusion framework that combines Convolutional Neural Networks (CNNs) and Long Short-Term Memory (LSTM) networks to enhance diagnostic performance. Our hybrid architecture processes ECG signals using LSTM, and extracts spatial features from thermal and X-ray images using separate CNN branches. An attention-based fusion module then integrates the features from each modality, enabling the model to prioritize the most informative aspects of the input data. Extensive evaluations on benchmark datasets demonstrate that the proposed method significantly outperforms unimodal and traditional fusion approaches in terms of accuracy, precision, recall, and AUC. Additional analyses, including modality-wise performance breakdown and ablation studies, further validate the effectiveness and robustness of our architecture in real-world medical diagnostic scenarios.