Decoding Parkinson's diagnosis: An OCT-based explainable AI with SHAP/LIME transparency from the Persian Cohort Study

Area of Science:

• Ophthalmology
• Neurology
• Artificial Intelligence

Background:

• Parkinson's disease (PD) diagnosis is challenging due to subjective assessments and late symptom onset.
• Retinal optical coherence tomography (OCT) offers non-invasive biomarkers for neurodegeneration.
• Integrating OCT with explainable AI (XAI) can improve PD diagnostic accuracy.

Purpose of the Study:

• To develop and validate an explainable AI framework for early Parkinson's disease diagnosis using retinal OCT biomarkers and clinical data.
• To identify key OCT and clinical features predictive of PD.
• To enhance diagnostic transparency and reliability.

Main Methods:

• A 6-layer deep neural network (DNN) was developed using OCT biomarkers (foveal thickness, volume) and clinical data (motor, olfactory) from the Persian Cohort Study.
• Synthetic Minority Oversampling (SMOTE) was employed to address class imbalance (PD:Healthy ≈ 1:5).
• SHAP and LIME were used for model interpretability, providing global and local feature explanations.

Main Results:

• The explainable AI framework achieved 95.3% accuracy and 0.98 AUC-ROC for PD diagnosis.
• Key biomarkers identified include SUPERIOR4 thickness (<120 µm) and foveal volume expansion (>0.15 mm³), alongside motor and olfactory deficits.
• SMOTE reduced false negatives by 12% while maintaining high specificity (94.8%).

Conclusions:

• This study presents a transparent, OCT-based AI framework for early PD detection via retinal neurodegeneration patterns.
• The multimodal, explainable, and robust model is suitable for resource-limited settings.
• Further validation across diverse populations and standardization of OCT protocols are recommended.