Development and Evaluation of a Method for Automated Detection of Spreading Depolarizations in the Injured Human Brain.

BACKGROUND: Spreading depolarizations (SDs) occur in some 60% of patients receiving intensive care following severe traumatic brain injury and often occur at a higher incidence following serious subarachnoid hemorrhage and malignant hemisphere stroke (MHS); they are independently associated with worse clinical outcome. Detection of SDs to guide clinical management, as is now being advocated, currently requires continuous and skilled monitoring of the electrocorticogram (ECoG), frequently extending over many days. METHODS: We developed and evaluated in two clinical intensive care units (ICU) a software routine capable of detecting SDs both in real time at the bedside and retrospectively and also capable of displaying patterns of their occurrence with time. We tested this prototype software in 91 data files, each of approximately 24 h, from 18 patients, and the results were compared with those of manual assessment ("ground truth") by an experienced assessor blind to the software outputs. RESULTS: The software successfully detected SDs in real time at the bedside, including in patients with clusters of SDs. Counts of SDs by software (dependent variable) were compared with ground truth by the investigator (independent) using linear regression. The slope of the regression was 0.7855 (95% confidence interval 0.7149-0.8561); a slope value of 1.0 lies outside the 95% confidence interval of the slope, representing significant undersensitivity of 79%. R2 was 0.8415. CONCLUSIONS: Despite significant undersensitivity, there was no additional loss of sensitivity at high SD counts, thus ensuring that dense clusters of depolarizations of particular pathogenic potential can be detected by software and depicted to clinicians in real time and also be archived.