Amblyopia, the condition often called lazy eye, may soon be diagnosed from eye movements alone. Researchers in Italy and the US trained an AI classifier on fixation patterns, targeting patients who struggle with standard vision tests.
Conventional diagnosis leans on subjective visual-acuity exams, which are unreliable in young children and in people with nystagmus. Mistakes lead to delayed treatment.
The study drew on 510 quantitative features from fixation recordings of 123 people, 86 of them amblyopic. Nine markers survived a genetic-algorithm search run alongside a neural network, covering both fast and slow eye-movement behavior. The classifier scored 100% accuracy on its test data; repeated 5-fold cross-validation averaged 85.8% accuracy with 94.1% specificity.
That robustness matters because deep learning normally demands huge datasets. This hybrid method works in small, high-dimensional settings, opening a route to objective screening.
The open-access paper ran August 17 in Scientific Reports, from Stefano Ramat and Damiano Virdis of the University of Pavia and Fatema Ghasia of the Cleveland Clinic’s Cole Eye Institute.
