Multi-modal deep learning framework for early detection of Parkinson's disease using neurological and physiological data for high-fidelity diagnosis.

Parkinson's disease (PD) is a progressive neurodegenerative disorder that remained challenging for proper diagnosis in its early stages due to its heterogeneous symptom presentation and overlapping clinical features. Consequently, there is no consensus on effectively detecting early-stage PD and classifying motor symptom severity. Therefore, the proposed research introduced MultiParkNet, an avant-grade multi-modal deep learning framework for early-stage PD detection synthesizing diverse neurological and physiological data sources. The proposed system integrated audio speech patterns, motor skills drawing characteristics, neuroimaging data, and cardiovascular signals with different neural architectures for robust feature extraction and fusion. The probabilistic classification approach enhanced disease identification with high fidelity and early detection. The model demonstrated exceptional performance, with an average training accuracy of 99.67%, validation accuracy of 98.15% [Formula: see text] and test accuracy of 96.74% [Formula: see text] across cross-validation experiments. This novel architecture significantly improved diagnostic precision with a transformative, AI-driven approach for Parkinson's disease assessment and potential clinical implications.