Deep Learning with Self-Attention Mechanisms for Parkinson’s Disease Classification Through MRI images

Parkinson’s disease (PD) is a prevalent neurodegenerative disorder affecting millions globally, characterized by the progressive loss of dopamine-producing neurons in the brain. Advanced machine learning and deep learning approaches, combined with medical imaging, have become instrumental in diagnosing PD. Magnetic resonance imaging (MRI), particularly T1-weighted scans, has proven effective in detecting structural brain changes linked to dopamine deficiency, making it a critical tool for PD diagnosis. This study focuses on developing and evaluating robust convolutional neural networks (CNNs) integrated with a self-attention mechanism for enhanced feature extraction and accurate PD classification. A Random Forest (RF) classifier differentiates PD patients and healthy controls (HC). The proposed approach achieves superior performance, with an accuracy of 95.78%, a sensitivity of 96%, and a specificity of 96%. This model outperforms other advanced techniques, making it a reliable tool for accurate diagnosis of PD and a helpful aid for neurologists in making clinical decisions.