Artificial intelligence applications in Parkinson's disease via retinal imaging

A comprehensive review paper by researchers including Ali Jafarizadeh and eight co-authors demonstrates that artificial intelligence applied to retinal imaging can serve as a powerful, non-invasive tool for detecting Parkinson's disease (PD). Analyzing 35 years of evidence from 19 qualifying studies, the research identifies three core AI diagnostic tasks: disease classification, retinal vessel segmentation, and PD risk stratification. The standout model, ShAMBi-LSTM, achieved a remarkable 97.2% accuracy, 99.5% precision, and a 0.981 F1 score for classifying PD on the Drishti dataset. This approach leverages the retina's microvasculature as a window into neurodegeneration, correlating structural changes with the progression of Parkinson's.

Beyond classification, the review highlights the nnU-Net model for segmentation with a 99.7% accuracy and a 98.9% Dice score, and AlexNet for risk prediction. The findings suggest that integrating AI with routine retinal scans could revolutionize PD screening by providing an objective, cost-effective, and accessible method for early detection. This is critically important as global PD cases are projected to rise sharply with an aging population. Early diagnosis via such a scalable tool could enable timely interventions to slow disease progression and significantly reduce long-term healthcare burdens, moving detection from the clinic to potentially standard eye exams.