Leveraging Eye-Tracking Signals for Neurodegenerative Disease Classification with Deep Learning Models.

Area of Science:

• Ophthalmology
• Neurology
• Artificial Intelligence

Background:

• Neurodegenerative diseases (NDs) cause subtle, detectable changes in eye movements.
• Accurate diagnosis of NDs, including Parkinson's Disease (PD) and Alzheimer's Disease (AD), is crucial for timely intervention.
• Distinguishing PD from PD-mimics (PDM) is a clinical challenge.

Purpose of the Study:

• To investigate the efficacy of deep learning models in classifying NDs using raw eye-tracking data.
• To identify which eye movement tasks are most effective for differentiating specific NDs.
• To assess the contribution of pupil size and binocular data to classification accuracy.

Main Methods:

• Collected eye-tracking data from 133 participants (51 healthy controls, 25 AD, 39 PD, 18 PDM) during smooth pursuit, text reading, and picture description tasks.
• Applied deep learning models to analyze raw eye movement data.
• Conducted ablation studies to evaluate the impact of pupil size and binocular data.

Main Results:

• The text reading task effectively distinguished healthy controls from PD and AD patients.
• The smooth pursuit task differentiated PD from PDM.
• Incorporating pupil size improved classification accuracy (especially for CTL vs. PD), and binocular data was vital for PD vs. PDM classification.

Conclusions:

• Deep learning models can accurately classify NDs using eye-tracking data, offering a potential objective evaluation tool.
• Eye movement analysis, particularly during specific tasks, shows promise for early diagnosis and differentiation of neurological disorders.
• This approach has clinical relevance for neurology and primary care settings.