Explainable AI for Early Detection of Neurological Disorders Using Multimodal Biomedical Data

Early and accurate diagnosis of neurodegenerative diseases is critical for guiding effective interventions, yet existing methods often rely on single‐modality data and offer limited interpretability. We introduce a unified, attention‐guided deep learning framework that fuses resting‐state functional MRI, electroencephalography, and cerebrospinal fluid biomarker assays to distinguish between Alzheimer’s disease and Parkinson’s disease in a cohort of 200 clinically diagnosed subjects. Modality‐specific encoders—a 3D convolutional neural network for fMRI, a bidirectional LSTM for EEG, and a fully connected network for CSF measurements—extract complementary feature embeddings, which are adaptively weighted through a learnable attention mechanism. Explainability is achieved via Grad‐CAM heatmaps localized on brain volumes and SHAP value analysis on the fused representation and molecular features, providing spatial, spectral, and biochemical insights that align with established neuropathological markers. Trained end-to-end with cross-entropy loss and regularized through data augmentation and early stopping, the proposed model attains classification accuracies of 92% for Alzheimer’s and 89% for Parkinson’s, outperforming unimodal baselines (78–84%) and classical SVM benchmarks (82%). Ablation studies confirm the benefit of attention fusion (+7% over the strongest single modality) and demonstrate that explainability components incur no performance penalty. These results underscore the potential of multimodal, interpretable AI to improve diagnostic precision and clinical trust.