University of Twente Student Theses


Using Artificial Intelligence to predict Intracranial Hypertension in traumatic brain injury patients

Willemse, I.H.J. (2020) Using Artificial Intelligence to predict Intracranial Hypertension in traumatic brain injury patients.

Full text not available from this repository.

Full Text Status:Access to this publication is restricted
Abstract:Objective: Prediction of intracranial hypertension in traumatic brain injury patients. We studied the potential to predict intracranial hypertension using the averaged intracranial pressure (ICP) wave and the use of a neural network and investigated the transfer function from arterial blood pressure (ABP) to ICP as a possible measure for cerebral autoregulation. Methods: 60 traumatic brain injury patients were included in this retrospective study. Patients were admitted to the intensive care unit (ICU) at RadboudUMC, Nijmegen. The ICP was monitored using an intraparenchymal probe. All patients were divided into three categories based on treatment intensity and ICP. A preprocessing method based on hierarchical clustering was implemented to detect and remove artefacts in the ICP signal. A convolutional neural network (CNN) was used to classify waves leading to intracranial hypertension (pre-IH) and intracranial hypertension (IH) waves from control waves. Furthermore, the CNN was used to classify control waves from patients of the three different categories. In addition, transfer function of ABP to ICP was calculated for two frequency bands with the associated phase, gain and coherence function and tested on patients with intact and impaired cerebral autoregulation. Results: ICP waves 60 minutes prior to IH can be distinguished from control waves with a sensitivity of 82%, a specificity of 97%, an accuracy of 88% and an AUROC of 0.97. An important clinically relevant finding is the ability of the model to classify control waves from the three different patient categories. The results of the transfer function analysis suggest that the gain component of the transfer function is a possible measure of cerebral autoregulation. Conclusion: We presented a deep learning method to classify IH and pre-IH waves from no-IH waves. Based on an averaged ICP waveform the CNN was able classify between ICP waves leading to hypertension and control waves based on the waveform morphology up to 1 hour in advance. There was no improvement in the classification performance of control waves when adding the gain of the transfer function.
Item Type:Essay (Master)
Faculty:TNW: Science and Technology
Subject:44 medicine
Programme:Technical Medicine MSc (60033)
Link to this item:
Export this item as:BibTeX
HTML Citation
Reference Manager


Repository Staff Only: item control page