Smartphone-based Pupillometry for Diagnosis of Ischemic and Hemorrhagic Stroke

Introduction Stroke is associated with significant morbidity and mortality [1]. The most common subtypes are acute ischemic (AIS) and intraparenchymal hemorrhage (IPH). It is important to discriminate between these stroke subtypes because they have radically different treatments. However, similarities in symptoms can make triage challenging [2, 3]. The current diagnostic test is computed tomography, but this test is expensive and only available in hospitals. A portable, inexpensive and easy-to-use method of distinguishing AIS from IPH would be useful. The pupillary light reflex (PLR) parameters are quantitative biomarkers of neurological status [4]. Pupillometry the method of obtaining these parameters. We studied the utility of quantified PLR parameters in patients with AIS and IPH using a mobile smartphone-based digital pupilometer [5]. Methods IPH subjects were recruited within 4 days of onset and AIS subjects were recruited before undergoing treatment. PLR parameters included: maximum diameter (MAX), minimum diameter (MIN), percent change (CHANGE), latency (LAT), mean constriction velocity (MCV), maximum constriction velocity (MAXCV), and mean dilation velocity (MDV). Random forest, k-nearest neighbors, support vector machine, and logistic regression binary classification models with a k=10 fold cross validation technique were trained. Models were evaluated for accuracy, sensitivity, specificity, area under the curve (AUC), and F1 score. Hyperparameters were not altered. Results The n=15 IPH patient recordings were from n=11 patients 23% female, mean age 58 years. The n=22 AIS recordings were from n=22 patients 59.1% female, mean age 68.9 years, and 77.3%. The best-performing model was support vector machine combination of CHANGE, MAX, MIN, and DV parameters (80% accuracy, 80% sensitivity, 85% specificity, AUC 0.83, and F1 score 0.78). Conclusion PLR parameters may be of use in classification of AIS versus IPH to assist clinical decision-making and triage of stroke patients. This could improve the timeliness of interventions, patient outcomes, morbidity and mortality.