UNLEASHING THE SYNERGY OF MULTIMODAL CONVOLUTIONAL RECURSIVE DBN OF PARKINSON'S DISEASE SEVERITY PREDICTION

Abstract

Parkinson's disease (PD) is a progressive neurodegenerative disorder characterized by the gradual loss of dopamine-producing neurons in the brain. PD typically affects individuals over the age of 60, although early-onset cases exist. Existing methods for diagnosing Parkinson's disease often rely on clinical assessments, including the Unified Parkinson's Disease Rating Scale (UPDRS), neuroimaging techniques like MRI and PET scans, and genetic analysis. While these methods have been valuable, they have several drawbacks: they can be subjective, leading to variability in diagnosis, they are resource-intensive, expensive, and may not detect early-stage PD effectively. Moreover, the reliance on single modalities limits the comprehensive understanding of the disease, necessitating the development of more accurate, accessible, and multimodal approaches for diagnosis and monitoring. In this study, we present a novel multimodal deep learning framework for predicting the severity of Parkinson's disease (PD), integrating Convolutional Neural Networks (CNNs) for feature extraction and Recurrent Neural Networks with Deep Belief Networks (RNN-DBN) for feature selection. Leveraging clinical assessments, imaging, and genetic data, our CNNs efficiently capture spatial and temporal patterns within each data modality while preserving inter-modal relationships. Subsequently, our RNN-DBN architecture adeptly exploits temporal dependencies, facilitating a deeper understanding of PD symptom evolution and enhancing the interpretability of the model. Evaluation on a diverse PD dataset demonstrates superior predictive performance, making our approach a valuable tool for clinicians to assess disease severity, contributing to more effective diagnostics and monitoring for Parkinson's disease.

Keywords: Parkinson's disease (PD), Convolutional Neural Networks (CNN), Deep Belief Networks (DBN), Rat Swarm Optimization (RSO).