Eye movement detection using electrooculography and machine learning in cardiac arrest patients.

Aim: To train a machine learning algorithm to identify eye movement from electrooculography (EOG) in cardiac arrest (CA) patients. Neuroprognostication of comatose post-CA patients is challenging, requiring novel biomarkers to guide decision making. Eye movement may be a promising marker of arousal recovery, as pathways for eye movement and arousal share common anatomic structures. Continuous quantification of eye movement is feasible through electroencephalogram (EEG) with EOG, but manual quantification is resource-intensive.

Methods: We conducted a retrospective, single-center cohort study of post-CA patients who underwent standard-of-care EEG/EOG monitoring in the intensive care unit from 2020 to 2023. We trained a machine learning algorithm to detect eye movement on one-hour of EOG data from 145,800 one-second samples from 48 patients. Performance was assessed on a reserved test set of 12-hours of EOG data from 705,600 one-second samples from 24 patients using area under the curve (AUC), sensitivity, and specificity.

Results: Of 72 eligible patients, average age was 56.9 years, and 46 (63.9%) were female. In the training group of 48 patients, 35 (72.9%) survived and 32 (66.7%) followed commands. In the test group, 16 (66.7%) survived and 7 (29.2%) followed commands. Our final algorithm identified eye movement with sensitivity of 94.0%, specificity of 82.0%, and an AUC of 94.2%.

Conclusion: Automated eye movement detection from EOG is highly sensitive in CA patients. Potential applications include using eye movement quantification to evaluate associations with recovery.