Machine Learning-Based Visuospatial and Eye-Movement Tasks for Early Detection of Lewy Body Dementia in Older Adults

Abstract

Detecting Lewy Body Dementia (LBD) early can be difficult because visuospatial (how well you can see where things are) and executive (how well you can organize things) problems happen before there are obvious symptoms of LBD. This study introduces a computer-based model to detect early indicators of LBD in adults aged 60 years or older by evaluating performance on tasks involving visuospatial processing and eye movement. Three types of tasks were used to determine LBD risk: a baseline gaze task measuring natural eye behavior (e.g., blink rate, head stability), and a memory task requiring recall of object locations using a 4 × 4 (16-cell) spatial grid AML, and an inhibition task where the person was required to have a structured way to stop their eyes from doing an involuntary movement when looking in one direction and then move their eyes in the opposite direction in order to create a new eye movement. When completing the LBD test tasks, subjects were required to use a tablet. Also, the test subjects' eyes were tracked in real-time using the video camera built into the tablet and data from their eye movements were recorded and analysed. In addition to eye behaviour data, tests were given to obtain behavioural data. A statistical model based on logistic regression was created using the findings of both eye behaviour and behavioural measures that predicted whether or not an LBD diagnosis would be made to create a summary score of each participant’s risk for developing LBD. The results from this study demonstrated how incorporating both visuospatial task performance and eye movement analysis provides evidence for the detection of subtle cognitive difficulties and that this is an effective method for identifying early dementia without requiring invasive procedures or time-intensive assessments.