Machine learning for comprehensive prediction of high risk for Alzheimer’s Disease based on chromatic pupilloperimetry

Abstract

Purpose : To characterize the pupil light reflex (PLR) for small focal chromatic light stimuli in 125 cognitively normal middle-aged subjects at high risk for Alzheimer’s disease (AD), due to family history and 61 controls.

Methods : 186 subjects were enrolled, 125 offspring of AD patients (FH+) and 61 age-matched controls (FH-), ages 44-71. Ophthalmic assessments included a Chromatic Pupilloperimetry test and a complete ophthalmologic examination to exclude ocular pathologies. Cognitive assessment, to verify subjects were asymptomatic, included executive function and episodic memory tests. 35 pupilloperimetry features were measured in 54 spots in a 24-2 visual field. Machine learning classification models were trained such that each model was introduced to a single feature type (measured in 54 spots) in the training data, and the same hyperparameters and training protocol were used in all models. Each model was then tested to quantify how well it can discriminate between FH+/FH-. The accuracy of a model was used as an indication of the correlation of a feature type to AD family history, using a standard confidence interval (CI) of 95%.

Results : Chromatic pupilloperimetry-based learning models were highly discriminative in differentiating subjects with and without AD family history, using short focal red (primarily cone-mediated), and dim blue (primarily rod-mediated) light stimuli. Features associated with transient PLR latency achieved Area under the Receiver Operating Characteristic Curve (ROC AUC) of 0.90 ± 0.051 (left-eye) and 0.87 ± 0.048 (right-eye). Parameters associated with the contraction arm of the transient PLR were more discriminative compared to parameters associated with the relaxation arm.

Conclusions : Chromatic pupilloperimetry differentiated between individuals at high risk for AD due to a parental family history from those without AD family history, with high sensitivity and specificity. Subtle changes in pupil contraction latency may be detected decades before the onset of AD clinical symptoms using a simple, non-invasive test.

This abstract was presented at the 2022 ARVO Annual Meeting, held in Denver, CO, May 1-4, 2022, and virtually.