Segment anything model-based segmentation with inception-ResNet-v2 classifier for Parkinson's disease diagnosis and...

Parkinson's disease (PD) is a neurological condition that distresses brain cells and can severely impact a person's healthy life. Accurate diagnosis of PD is vital for early intervention. However, many existing works commonly use Voxel-Based Morphometry for brain analysis, which requires rigid preprocessing steps such as spatial normalization and template alignment that introduce biases and limit flexibility in segmenting complex or irregular brain structures. The study aims to improve PD prediction by addressing the limitation of the VBM technique by introducing the advanced automated segmentation method Segment Anything Model (SAM) to enable more accurate and reliable diagnosis. Additionally, the study presents a metaverse framework to provide a better understanding and decision-making in PD diagnosis. The proposed model employs advanced pre-processing procedures such as denoising, brain part separation, motion correction, and data augmentation to improve the brain 3D MRI image quality. The filtered images are then segmented using the SAM model. Features are extracted from segmented images using linear binary patterns, Gray-level co-occurrence motion, and shape features to enhance the robustness of PD classification. We use the Success History Intelligent Optimizer algorithm to hand-pick the optimal features for better diagnosis to avoid the local minima. The selected structures are classified with Inception-ResNet-v2 to fix the presence of PD. The proposed model demonstrates superior performance across various performance metrics for PD diagnosis. The proposed method accomplishes an accuracy of 99.75 % in predicting PD, highlighting its performance in accurately classifying PD and leading to improved patient outcomes.