Diagnosing Neurourological Complications of Parkinson’s Disease with Self-Adaptive Convolutional Neural Networks through MRI Analysis

Abstract

Parkinson's disease (PD) is a progressive and degenerative neurological disorder that primarily disrupts the brain's motor functions, leading to symptoms like bradykinesia, stiffness, balance problems, and resting tremors. The intricate nature of PD, which often resembles other neurological conditions and involves subtle structural brain changes, complicates accurate diagnosis, resulting in a 25% diagnostic error rate. To address this challenge, the research community has employed various machine learning techniques using manually crafted features for diagnosis. This study introduces an innovative computer-aided diagnostic approach for PD based on a self-adaptive convolutional neural network (SCNN), a potent model for automatically extracting essential problem features. The Parkinson's Progression Markers Initiative (PPMI) provided the dataset used in this investigation, which includes a variety of datasets such as T2-weighted MRI scans from both PD individuals as well as healthy controls (HC). In particular, mid-slice MRI data is gathered and registered in order to be aligned. To pinpoint the region of interest in the midbrain, a 33 × 33-sized window is employed, as PD primarily affects the substantia nigra within the midbrain. Comprehensive experiments have been conducted to validate the reliability of the SCNN approach. Using common evaluation measures like area under the curve, specificity, sensitivity, and accuracy, the suggested method's performance is evaluated. Notably, the evaluation findings show that in terms of diagnostic precision, the SCNN performs better than other machine learning techniques