Automated Alzheimer’s, Mild Cognitive Impairment, and Normal Aging Screening using Polar Transformation of Optic Disc and Central Zone of Fundus Images

Detecting Mild Cognitive Impairment (MCI) is crucial for mitigating the risk of Alzheimer’s disease (AD), a leading global cause of death. However, the current gold standard for AD and MCI detection relies on specialized equipment often limited to large testing centers, particularly in low-resource settings like Thailand. Our previous work aimed to create a cost-effective MCI and AD screening method using fundus images but struggled to differentiate between AD and MCI. Henceforth, we developed the proposed methodology, utilizing DenseNet-121 on polar-transformed and zone-selected fundus images, which significantly enhances AD and MCI classification, achieving 83% accuracy, 90% sensitivity, 77% specificity, 87% precision, and an F-1 score of 88%. Moreover, the model’s Grad-Cam++ heatmap highlights vasculature differences, particularly in tortuosity and thickness, between AD and MCI fundus images. Combined with our previous work, we created a fully automated pipeline model for MCI, AD, and Normal aging classification, which is inexpensive, fast, and non-invasive with an overall 3-class accuracy of 88%.