Toward learning intracranial hypertension through physiological features: A statistical and machine learning approach

Intracranial pressure (ICP) monitoring is one of the most routinely conducted procedures in neurocritical care units to monitor comatose patients having Traumatic Brain Injuries (TBI). The existing clinical standard of care using insertion of a transducer or catheter is based on an invasive method which always has certain risks. This study found some significant correlations between ICP and routinely monitored physiological signals in TBI patients. The results indicate that Heart Rate, Pulse, Diastolic Arterial Blood Pressure, Respiration, Mean Arterial Blood Pressure and ECG ST segment levels are significantly correlated with ICP and have the potential to be reliable predictors of intracranial hypertension (ICH). In this paper an algorithm is presented to extract increased ICP episodes and the corresponding physiological signals from large datasets. Episodes of “Intracranial hypertension onset”, “Intracranial Hypertension” and “Severe Intracranial Hypertension” were detected and differentiated based on the ICP levels. It was discovered that there is a potential predictive power in the dynamical information of routinely monitored physiological signals in TBI patients.