MultimodalCNN-PD: a Parkinson's disease diagnostics framework using multimodal convolutional neural network.

Area of Science:

• Artificial Intelligence in Medicine
• Neuroimaging and Computational Neuroscience
• Biomedical Data Science

Background:

• Parkinson's disease (PD) is a common neurodegenerative condition impacting motor and cognitive abilities.
• Early detection, especially during the prodromal phase, is crucial for timely intervention and management.

Purpose of the Study:

• To develop and validate a deep learning model (MultimodalCNN-PD++) for enhanced classification of Parkinson's disease.
• To integrate multimodal data, including MRI and clinical metadata, for improved diagnostic accuracy.

Main Methods:

• Utilized a deep learning architecture with EfficientNetB0, Mobile CBAM, and MGCA++.
• Implemented hierarchical feature selection for clinical data and BioClinicalBERT with LoRA for metadata processing.
• Integrated Magnetic Resonance Imaging (MRI) with comprehensive clinical metadata.

Main Results:

• Achieved 97.5% accuracy on the PPMI dataset for classifying normal controls, prodromal PD, and diagnosed PD.
• Demonstrated robust generalizability with 96.2% accuracy on the external OASIS-3 dataset.
• Confirmed the significant contributions of individual model components through ablation studies.

Conclusions:

• The MultimodalCNN-PD++ framework establishes a new benchmark for multiclass PD diagnosis.
• The model shows potential as a clinically deployable AI tool for early detection and personalized management of neurodegenerative diseases.