Validation and Comparison of Pupillary Reflex Indices in TBI and Matched Healthy Controls Using PupilScreen, a Smartphone-based Pupillometer

INTRODUCTION: Pupillary light reflex (PLR) is one of the earliest biomarkers used to assess, diagnose, and treat neurological conditions, particularly traumatic brain injury (TBI). PLR indices have been examined as an indicator for specific neurological conditions using a pupillometer. We developed an alternative, mathematically driven stand-alone app to increase precision, to provide broader accessibility, and utilization of automated pupillometry. METHODS: The second-generation PupilScreen app was developed using a fully convolutional neural network (FCNN) model, using the standard iPhone light source and camera without any external environmental controls. We examined n=21 TBI subjects with n=21healthy controls matched to primary eye color. We analyzed known pupillary biomarkers for TBI and compared the metrics in both cohorts. Additionally, we examined age, gender, ethnicity, GCS, eye color, and primary lesion. RESULTS: Distribution of eye color was 38% brown eyes, 29% blue eyes, and 33% mixed eyes in both cohorts. Mean age was 54 vs 35 in TBI and healthy controls, respectively. Female sex was 19% versus 62% TBI and healthy controls sequentially. Ethnicities included White (TBI 67%, healthy controls 76%), African American (TBI 9%, healthy controls 10%), Asian (TBI 9%, healthy controls14%), Native American/Indian (TBI 5%, healthy controls 0%). Lesions consisted of subdural hematoma (61%), intraparenchymal hemorrhage (24%), contusions (10%), traumatic subarachnoid hemorrhage (5%), and 38% of lesions were multifocal. Mean GCS was 8, and 52% were classified as severe TBI. PupilScreen measurements included mean latency of 0.45s(L)/0.31s(R) in TBI and 0.17s (L/R) (p<0.05) healthy controls; mean % diameter change was 21%(L)/17%(R) in TBI, 39%(L/R) (p<0.05) in healthy controls; mean constriction velocity was 3.7px/s(L)/3.6px/s(R)in TBI and 11.4px/s(L)/11.1px/s(R) (p<0.05) n healthy controls; and the mean dilation velocity was 2.4px/s(L/R) and 4.1px/s(L)/4.5px/s(R) (p<0.05) in TBI and healthy cohorts respectively. CONCLUSION: The preliminary PupilScreen results demonstrated capability in identification of PLR biomarkers in TBI population compared to healthy cohort without the need for external environmental controls. Further development of PupilScreen app is warranted.