Machine Learning for Early Parkinson's Disease Identification within SWEDD Group Using Clinical and DaTSCAN SPECT...

Area of Science:

• Neurology
• Medical Imaging
• Artificial Intelligence

Background:

• Early diagnosis of Parkinson's Disease (PD) is crucial for patient management.
• The Scan Without Evidence of Dopaminergic Deficit (SWEDD) group presents diagnostic challenges due to heterogeneous clinical and imaging features.
• Machine Learning (ML) offers potential for distinguishing PD patients from Healthy Controls (HC) within SWEDD.

Purpose of the Study:

• To evaluate the efficacy of ML algorithms in differentiating PD patients from HC within the SWEDD cohort.
• To compare the performance of Principal Component Analysis (PCA) and Linear Discriminant Analysis (LDA) for data reduction.
• To assess the diagnostic capabilities of Density-Based Spatial (DBSCAN), K-means, and Hierarchical Clustering models.

Main Methods:

• Analysis of data from 548 subjects using PCA and LDA for dimensionality reduction.
• Application of DBSCAN, K-means, and Hierarchical Clustering models using LDA results.
• Performance evaluation of clustering models against ground truth based on accuracy, sensitivity, and specificity.

Main Results:

• Linear Discriminant Analysis (LDA) outperformed Principal Component Analysis (PCA).
• Hierarchical Clustering achieved superior performance compared to DBSCAN and K-means.
• Hierarchical Clustering demonstrated significant improvements in accuracy (64%), sensitivity (78.13%), and specificity (38.89%) over other models.

Conclusions:

• ML models are suitable for distinguishing PD patients from HC within the SWEDD group.
• Hierarchical Clustering, combined with LDA, shows promise as a diagnostic tool for early PD detection in challenging cases.
• This approach aids in accurate patient stratification and treatment planning for individuals with potential PD.