Machine Learning with Objective Serum Markers and Algorithmic Deep Learning Computed Tomography Scan Analysis for Classification of Brain Injury

Abstract

Brain injury is pathophysiologically diverse, with many cases presenting with mixed pathologies. Utilizing serum biomarkers to investigate the pathophysiology of injury would help to aid in understanding prognosis and targeting therapeutics. One goal of the study is to develop a traumatic brain injury classification scheme based on two serum biomarkers glial fibrillary acidic protein (GFAP) and ubiquitin carboxy-terminal L1 (UCH-L1). GFAP and UCH-L1 serum marker analysis was performed on patients with isolated traumatic brain injury or healthy, uninjured controls within 32 hours of hospital admission. Machine learning was utilized for classification of brain injury and to develop a novel algorithm capable of classifying the type of brain injury based on GFAP and UCH-L1 concentrations. Each patients brain injury was classified using standard clinical and radiographic assessments and stratified into one of four trauma groups: trauma, spontaneous hemorrhage, oxygen deprivation, or a high-velocity trauma with negative radiographic finding. Analysis of prospectively collected serum for GFAP and UCH-L1 was performed on 61 patients and 39 controls. The subjects with trauma, spontaneous hemorrhages and oxygen deprivation could be distinguished from controls with AUC = 1.00. Combination of GFAP and UCH-L1 concentrations distinguished the high-velocity injuries that were negative for radiographic indicators (CT-negative) from controls with AUC of 0.93. Serum biomarker profiles were found to accurately predict etiology across four distinct brain injuries, including CT-negative. Serum markers GFAP and UCHL1 may be helpful for classifying the nature of brain injury, which will aid with prognostication and development of therapeutics.