Diagnosis and Prognosis Perspective of Parkinson's Disease Using Multimodal Artificial Intelligence

Abstract

Parkinson's disease (PD) demands both accurate early diagnosis and reliable assessment of progression, yet its clinical manifestations are heterogeneous and often become evident only after substantial neurodegeneration has occurred. This paper investigates whether multimodal artificial intelligence (AI) can address these two complementary requirements within a unified analytical perspective, integrating neuroimaging, clinical, genetic, speech, and handwriting information. Adopting a pragmatist, design-science methodology, the study evaluates two architectures: an attention-based multimodal diagnostic framework (PMMD) fusing neuroimaging, speech, and handwriting features for early detection, and a hybrid CNN–LSTM–genetic framework for disease-stage classification and progression prediction. Diagnostic evaluation yielded 92% sensitivity, 90% specificity, 91% accuracy, and an area under the curve (AUC) of 0.94, outperforming reported unimodal baselines. Prognostic evaluation achieved 89.7% accuracy and 0.90 AUC for PD-stage classification, and progression prediction reached a mean squared error (MSE) of 0.021, root mean squared error (RMSE) of 0.145, and mean absolute error (MAE) of 0.112. These findings indicate that multimodal AI provides a broader, patient-level representation of PD by combining complementary diagnostic and temporal information: attention-based fusion supports early identification, while CNN–LSTM–genetic integration extends the framework toward staging and trajectory estimation. The paper contributes an integrated diagnosis–prognosis framework, a comparative empirical characterisation against unimodal approaches, and a roadmap toward personalized monitoring and clinical decision support. Limitations include reliance on reported and cross-sectional evaluation, modest and potentially non-representative cohorts, and limited external validation. The work advances neuro-informatics practice by demonstrating that diagnosis and prognosis need not be treated as isolated AI tasks, and by identifying the conditions—explainability, missing-modality handling, and prospective validation—under which multimodal systems can move toward precision PD management.