Multi-Class Deep Learning Architecture for Differential Diagnosis of Parkinson's Disease and SWEDD Using T2-Weighted MRI Features

Parkinson's disease is a progressive neurodegenerative disease that mainly affects the elderly and is caused by an ongoing degeneration of dopaminergic neurons in the substantia nigra. As symptoms worsen over time and global aging populations increase, the disease poses growing financial burdens on healthcare systems. With no definitive cure available post-onset, early detection is crucial for managing neuronal deterioration and maintaining patient quality of life. This study analyzed MRI T2 images from the PPMI database, examining 430 cases including 87 healthy subjects, 307 Parkinson's patients, and 36 SWEDD cases. After data preprocessing, a Deep Neural Network was developed to classify subjects into these three categories. With 93% accuracy, 0.93 recall, 0.91 precision, 0.92 F1-score and ROC-AUC of 0.94, 0.95, and 0.98 in Healthy, Parkinson's, and SWEDD classifications respectively, the model performed extremely well.