Title : Decoding parkinson’s diagnosis: An OCT-based explainable AI with SHAP/LIME transparency from the persian cohort study
Abstract:
Background: Parkinson’s disease (PD) remains challenging because of subjective clinical assessments and late-stage symptom manifestation. Retinal optical coherence tomography (OCT) biomarkers reflect neurodegenerative changes and provide a noninvasive diagnostic tool. This study integrated retinal OCT with explainable artificial intelligence (XAI) to improve PD diagnosis.
Methods: Using data from the Persian Cohort Study (204 patients with PD and 972 controls), we developed a 6-layer deep neural network (DNN) combining OCT biomarkers (foveal thickness and volume) and clinical variables (motor scores and olfactory dysfunction). Synthetic Minority Oversampling (SMOTE) mitigated class imbalance (PD:Healthy ≈ 1:5). Model interpretability was ensured through global feature importance (SHAP) and local explanations (LIME).
Results: An explainable AI framework integrating retinal OCT biomarkers and clinical data was used to diagnose PD with 95.3% accuracy and 0.98 AUC-ROC. The model identified SUPERIOR4 thickness reduction and foveal volume expansion as key biomarkers alongside motor/olfactory deficits. SHAP/LIME revealed interpretable thresholds (e.g., SUPERIOR4 <120 µm = high risk), whereas SMOTE reduced false negatives by 12% while maintaining specificity (94.8%).
Conclusion: This study pioneered a transparent OCT-based AI framework for PD diagnosis, emphasizing early detection through retinal neurodegeneration patterns. The integration of multimodal data, explainability, and imbalanced robustness makes it a scalable tool for resource-limited settings. Future studies should validate biomarkers across diverse populations and standardize OCT protocols.
Keywords: Parkinson’s Disease, Retinal Biomarkers, OCT, Deep Learning, Explainable AI
