Machine learning based prediction of imminent ICP insults during neurocritical care of traumatic brain injury

• Background/ObjectiveIncreased intracranial pressure (ICP) is a feared secondary brain insult in neurointensive care (NIC) of traumatic brain injury (TBI). A system that predicts ICP insults before they emerge may facilitate early optimization of the physiology, which in turn may lead to that the predicted ICP insult will never occur. The aim of this study was to evaluate the performance of different AI models in predicting risk of ICP insults. MethodsThe models were trained to predict risk of ICP insults starting within 30 minutes, using the Uppsala High Frequency TBI (UHF-TBI) dataset. A restricted dataset consisting of monitoring data only was used, and an unrestricted dataset using monitoring data as well as clinical data, demographic data and radiological evaluations. Four different model classes were compared: Gaussian Process Regression (GP), Logistic Regression (LR), Random Forest classifier (RF) and Extreme Gradient Boosted Decision Trees (XGBoost).ResultsSix hundred and two TBI patients were included (total monitoring 138 411 hours). On the task of predicting upcoming ICP insults, the GP model performed similar on the UHF-TBI dataset (sensitivity 93.2% and specificity 93.9%), as in earlier smaller studies. Using a more flexible model (XGBoost) resulted in slightly better performance (sensitivity 93.8% and specificity 94.6%). Adding more clinical variables and features further improved the performance of the models slightly (XGBoost: sensitivity 94.1% and specificity of 94.6%). Using AUROC as performance measure, the XGBoost models also performed slightly better than the other models. ConclusionsAI models have potential to become valuable tools for prediction of ICP insults in advance during NIC. The fact that common off-the-shelf models, such as XGBoost, performed well in predicting ICP insults opens for new possibilities, which can lead to faster advances in the field and faster clinical implementations.

Ämnesord

TEKNIK OCH TEKNOLOGIER  -- Elektroteknik och elektronik -- Signalbehandling (hsv//swe)

ENGINEERING AND TECHNOLOGY  -- Electrical Engineering, Electronic Engineering, Information Engineering -- Signal Processing (hsv//eng)

MEDICIN OCH HÄLSOVETENSKAP  -- Klinisk medicin -- Neurologi (hsv//swe)

MEDICAL AND HEALTH SCIENCES  -- Clinical Medicine -- Neurology (hsv//eng)

Publikations- och innehållstyp

vet (ämneskategori)

ovr (ämneskategori)