Deep learning predicts prevalent and incident Parkinson’s disease from UK Biobank fundus imaging

AI Generated Summary*

Fundus photographs from the UK Biobank were used to test whether machine learning can separate Parkinson's disease from healthy controls. After quality screening, 123 PD images from 84 participants were matched by age and gender to equal numbers of controls, with subsets for prevalent (77 images) and incident (46 images) disease. Four conventional classifiers and five convolutional networks were compared using repeated five-fold cross-validation. AlexNet did best, with AUC 0.77 and 68% accuracy overall, sensitivity 0.76 and specificity 0.60, plus AUCs of 0.73 for prevalent and 0.68 for incident cases; the strongest conventional model overall, an RBF support vector machine, reached AUC 0.71. Attribution maps aligned with segmented retinal structures and a marked fovea, and AlexNet resisted perturbations best.

*This summary was generated by AI and is published unedited. Oku does not alter these summaries. It may contain errors or omissions and is provided for general informational purposes only. Oku does not guarantee its accuracy, completeness, or reliability. For authoritative information, please refer to the original, peer-reviewed article.