Validation of reading as a predictor of mild cognitive impairment

Mild cognitive impairment (MCI) is a neurocognitive disorder that precedes Alzheimer's disease, but also other types of dementia. The use of reading tasks, when paired with eye-tracking technology, has been suggested as an effective biomarker for identifying MCI and distinguishing it from healthy individuals. The objective of this study was twofold: (1) to explore the disparities in eye movements during reading between individuals with MCI and healthy controls and train a predictive model to detect MCI, and (2) to validate these findings on a large independent dataset. We developed features for a model designed to automatically detect cognitive impairment based on the data of 115 subjects; 62 cognitively impaired and 53 healthy controls. Each subject was subjected to a neurological evaluation, a thorough psychological analysis, and completed a brief reading exercise while their eye movements were monitored using an eye-tracker. Their eye movements were characterised by patterns of saccades and fixations and were analysed across both groups. Several characteristics showed very high statistical significance, indicating differences in gaze behaviour between the groups. These characteristics were then employed to develop a machine learning model that differentiates cognitively impaired individuals from healthy controls. For the validation purposes, we ran a separate study with 99 new subjects using the same experimental design. The model reached about 75% AUROC. These results confirm that reading tasks can serve as a basis for early detection of MCI; however, complementary eye-tracking tasks are needed to further increase the detection accuracy.