Augmenting Dementia Cognitive Assessment with Instruction-less Eye-tracking Tests

Daniele Ravi

Abstract: Eye-tracking technology offers promise for enhancing dementia screening. This work introduces a novel approach for extracting salient features from an instruction-less eye-tracking cognitive test administered to patients with various dementia subtypes and healthy controls. We employ self-supervised representation learning, using cognitive activity recognition as a pretext task, to automatically learn discriminative features from raw eye-tracking sequences. The features extracted through this deep learning approach are more sensitive than traditional handcrafted eye-tracking metrics in detecting performance differences between participants with and without dementia across multiple cognitive tasks. Our results demonstrate that instruction-less eye-tracking tests can effectively detect abnormal oculomotor biomarkers associated with dementia-related cognitive dysfunction, providing a brief and low-stress assessment tool for dementia screening and monitoring.