Exploring the Potential Imaging Biomarkers for Parkinson's Disease Using Machine Learning Approach.

Area of Science:

• Neuroscience and Medical Imaging
• Artificial Intelligence in Healthcare

Background:

• Parkinson's disease (PD) involves dopamine neuron loss, impacting motor and neuropsychiatric functions.
• Dopamine transporter scans (DATSCAN) using SPECT assess dopaminergic neuron integrity.

Purpose of the Study:

• To develop a machine learning (ML) algorithm for PD diagnosis using DATSCAN data.
• To identify a key biomarker from DATSCAN images for improved diagnostic accuracy.

Main Methods:

• Trained an AdaBoost classifier on 13 DATSCAN parameters and handedness from 1309 individuals (PPMI database).
• Utilized Local Interpretable Model-Agnostic Explainer (LIME) for biomarker identification and model interpretability.

Main Results:

• Achieved 98.88% accuracy and 99.81% AUC with the ML model.
• Identified contralateral putamen SBR as the most significant predictive feature for PD diagnosis.

Conclusions:

• The ML approach simplifies PD diagnosis by focusing on a single, interpretable biomarker.
• This enhances diagnostic precision and supports clinical decision-making, despite DATSCAN's limitations in early-stage detection.