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

Area of Science:

• Neurology
• Medical Imaging
• Artificial Intelligence

Background:

• Accurate Parkinson's disease (PD) diagnosis is crucial for timely intervention.
• Traditional Voxel-Based Morphometry (VBM) for brain analysis has limitations, including preprocessing biases and inflexibility with complex brain structures.
• Existing diagnostic methods require improvement for enhanced accuracy and reliability.

Purpose of the Study:

• To enhance Parkinson's disease (PD) prediction accuracy by overcoming VBM limitations.
• To introduce the automated segmentation method Segment Anything Model (SAM) for more reliable PD diagnosis.
• To present a metaverse framework for improved understanding and decision-making in PD diagnosis.

Main Methods:

• Utilized advanced preprocessing techniques including denoising, brain part separation, motion correction, and data augmentation on 3D MRI images.
• Employed the Segment Anything Model (SAM) for automated segmentation of brain structures.
• Extracted features using linear binary patterns, Gray-level co-occurrence matrices, and shape descriptors.
• Optimized feature selection using the Success History Intelligent Optimizer algorithm.
• Classified PD using the Inception-ResNet-v2 model.

Main Results:

• The proposed model demonstrated superior performance in PD diagnosis across multiple metrics.
• Achieved a high accuracy rate of 99.75% in predicting Parkinson's disease.
• The integration of SAM and advanced feature extraction/selection significantly improved classification robustness.

Conclusions:

• The developed model offers a highly accurate and reliable approach for Parkinson's disease diagnosis using 3D MRI.
• The study highlights the potential of the Segment Anything Model (SAM) in neurological disorder analysis.
• The findings suggest improved diagnostic capabilities, potentially leading to better patient management and outcomes.