Neurodegenerative disorders, specifically Parkinson’s (PD) and Alzheimer’s disease (AD), present significant diagnostic challenges due to their heterogeneous symptomatology and overlapping clinical features. While deep learning has demonstrated remarkable potential in automated diagnosis, the "black box" nature of these models hinders their integration into clinical workflows, where interpretability is paramount.
This presentation introduces a comprehensive framework for Multimodal Explainable Artificial Intelligence (XAI) tailored to the early detection and progression monitoring of PD and AD. We utilize a [hybrid CNN-Transformer] architecture with a [late fusion] strategy to integrate diverse data modalities. By fusing neuroimaging [(MRI/PET)] and non-invasive biomarkers [(voice or gait analysis)], the proposed approach captures the multifaceted pathology of these diseases more effectively than unimodal systems. The model was evaluated on subsets of the public [ADNI and PPMI] cohorts, comprising approximately [1,200] subjects with paired multi-omic data.
Crucially, the framework integrates XAI techniques, including SHAP (SHapley Additive exPlanations) and attention mapping, to elucidate the algorithmic decision-making process. The clinical validity of these explainability outputs was strictly assessed through [expert neuroradiologist review and localization agreement] with established anatomical biomarkers. This allows clinicians to confidently visualize specific anatomical regions or signal features driving the model’s predictions. Empirical results demonstrate that our multimodal fusion achieves an AUC of [0.94] and an accuracy of [91%], outperforming standard unimodal baselines by [8-10%]. Ultimately, this work bridges the gap between high-performance AI and clinical trust, paving the way for reliable, personalized computer-aided diagnostic tools in neurology.