AI-Powered Deep Nomogram Model for Early Detection of Cognitive Impairment in Parkinson's Disease

Abstract

Cognitive impairment is a prevalent and debilitating non-motor symptom of Parkinson's Disease (PD), often emerging in its early stages and progressively worsening over time. Timely detection is critical to initiate personalized interventions and improve patient outcomes. This study presents a novel AI-powered deep nomogram model that integrates clinical, demographic, and neuropsychological features to facilitate the early prediction of cognitive decline in PD patients. The model combines the interpretability of traditional nomograms with the predictive strength of deep learning architectures, leveraging a hybrid approach that includes feature embedding, attention mechanisms, and multi-layer perceptrons. Utilizing a well-curated dataset derived from longitudinal PD studies, the proposed model demonstrates superior performance in terms of accuracy, sensitivity, and area under the receiver operating characteristic curve (AUC), outperforming conventional statistical models and shallow learning classifiers. Furthermore, explainability tools such as SHAP (Shapley Additive Explanations) are employed to elucidate the contribution of individual variables, thereby enhancing clinical trust and adoption. The findings underscore the potential of AI-driven nomograms as powerful diagnostic tools for early cognitive impairment in PD, enabling proactive management strategies in neurodegenerative disease care.