University of Twente Student Theses

Login

Predicting intracranial hypertension in acute traumatic brain injury patients

Smit, J.J. (2020) Predicting intracranial hypertension in acute traumatic brain injury patients.

[img] PDF
1MB
Abstract:Objective: Prediction of intracranial hypertension in acute traumatic brain injury patients. We studied the potential to predict intracranial hypertension by morphological features of the ICP signal and by the use of a neural network. Methods: Fifty-three traumatic brain injury patients were included in this retrospective study. These patients were admitted to the ICU at RadboudUMC, Nijmegen. The ICP was monitored using an intraparenchymal probe. Spectral regression analysis was performed to robustly detect the ICP-subpeaks and calculate 27 morphological metrics of the ICP signal (MOCAIP). The spectral regression method required a test and training set. 7 experts of the RadboudUMC labelled subpeaks in averaged ICP waves. The performance and consistency in peak labelling of the experts was evaluated by the percentage of inconsistency and intra-and-intraclass correlation coefficients. By lack of a gold standard the labelled waves were used as a gold standard to train and evaluate the performance of the spectral regression algorithm. A Kruskal Wallis test was performed to test for significance in the MOCAIP metrics between control waves, waves leading to intracranial hypertension (pre-IH) and intracranial hypertension (IH) waves. A classic machine learning approach was used to test the potential to predict intracranial hypertension based on the MOCAIP metrics by a classification tree. A CNN and LSTM neural network were used to classify IH and no intracranial hypertension (no-IH) based on the original ICP-waves. Results: We found statistical significance in 11 metrics between the IH waves and pre-IH waves. 5 metrics were significant between ICP waves leading to hypertension and control waves. Waves leading to hypertension could be distuingished from control waves with a sensitivity of 89% and specificity of 71%. The predicitive power in classification of pre-IH waves at specific timing prior to IH (5, 10, 15, 20 minutes) based on MOCAIP metrics is still limited. The obtained results from the neural network showed an accuracy of 70%, 97% sensitivity and 66 % specifity to classify between IH and no-IH waves. These results were very disparate for the test and training set. Conclusion: Prediction of intracranial hypertension is promising for classification between ICP waves leading to hypertension and control waves based on the MOCAIP metrics.
Item Type:Essay (Master)
Clients:
Radboud UMC, Nijmegen
Faculty:TNW: Science and Technology
Subject:44 medicine, 50 technical science in general
Programme:Technical Medicine MSc (60033)
Link to this item:http://purl.utwente.nl/essays/80722
Export this item as:BibTeX
EndNote
HTML Citation
Reference Manager

 

Repository Staff Only: item control page