Multimodal Analysis for Early Parkinson’s Disease Detection: Integrating Voice Analysis and Brain MRI Analysis with Web-Based Deployment

Abstract

Parkinson's disease is a progressive neurodegenerative disorder affecting over 10 million people worldwide, characterized by motor dysfunction and voice alterations that manifest early in disease progression. Traditional diagnostic approaches rely on expensive neuroimaging and subjective clinical evaluation, limiting accessibility and early detection capabilities. This study presents a comprehensive multimodal framework combining voice biomarker analysis and brain MRI imaging for automated PD detection, encompassing nine distinct machine learning approaches and advanced convolutional neural networks. The voice analysis component employs Mel-Frequency Cepstral Coefficients extracted from audio recordings, evaluated across four neural network variants including batch normalization, residual connections, and LSTM architectures, alongside five traditional ML models. The MRI component utilizes a custom CNN architecture with four convolutional blocks, batch normalization, and dropout regularization, trained on brain imaging data from Kaggle repositories. A production-ready Flask web application enables real-time multimodal assessment with voice-only, image-only, and combined prediction modes, featuring risk stratification, patient management, and comprehensive evaluation through confusion matrix analysis. Results demonstrate superior performance of ensemble methods and batch-normalized architectures, achieving high accuracy across stratified cross-validation and robust generalization. The deployed system provides accessible, cost-effective screening tools positioning this framework as a valuable clinical decision support system for early PD detection in telehealth and resource-constrained environments.