Integrative GF and FFT for Neuro MRI-based Parkinson's Disease Identification
Research Paper
Abstract
In this paper, we propose an automated detection system for identifying Parkinson’s disorder using advanced image analysis techniques on brain MRI data. Our method focuses on segmenting the white matter (WM) in the brain to efficiently distinguish between healthy individuals and those with Parkinson's disease (PD). The research is structured around three key stages. First, the MRI PD dataset is divided into two cohorts: healthy individuals and those diagnosed with PD. In the second stage, image preprocessing techniques, namely scaling and binarization, optimize the subsequent segmentation process. The third stage involves the precise delineation of the super colliculus region using Otsu thresholding and morphological operations such as opening, dilation, and closing. Our system employs Gabor Filters (GF) and Fast Fourier Transform (FFT) to extract distinctive features from each group, which are fused during both training and testing to enhance detection accuracy. The efficiency of our approach is validated using a novel Parkinson's Disease Shuffle-Squeeze Network (PDSSNet) applied to a benchmark MRI PD dataset from the Parkinson's Progression Markers Initiative (PPMI). Our system achieves an accuracy of 95.51% when analyzing WM regions in brain MRI data. Comparative analysis shows the superior performance of our GF-FFT approach against other feature extraction methods, demonstrating its potential as a robust tool for Parkinson's disorder detection. This work presents a significant advancement in accurately identifying Parkinson's disorder through WM segmentation in brain MRI images.