University of Twente Student Theses

Login

Machine Learning for CT-based clinical triage in hemorrhagic and ischemic stroke

Kuppens, D. (2018) Machine Learning for CT-based clinical triage in hemorrhagic and ischemic stroke.

[img] PDF
63kB
Full Text Status:Access to this publication is restricted
Abstract:Stroke is a 'brain attack'. Stroke occurs when blood ow to an area is disrupted, either by bleeding or by blockage of an artery. The aim of this thesis is to explore to which extent the clinical management of both hemorrhagic and ischemic stroke patients can become more informed by using machine learning techniques to both extract relevant features from more primary data and characterize the state of the patient's stroke. We assess the utility of quantitative features of contrast medium distribution and texture analysis of the hematoma in predicting hematoma expansion. Our model outperforms the conventional spot sign on the train set and outperforms the conventional spot sign on the test set in terms of specificity, but not in terms of sensitivity. We explore the use of deep learning to predict lesion shape and size using multiple CT acquisitions. We have designed a deep convolutional neural network (CNN), transforming a CT input into a probability distribution of the pixels belonging to an ischemic lesion. Our network overfitted on the training data, and therefore was not able to generalize to new, unseen data.
Item Type:Essay (Master)
Clients:
Massachusetts General Hospital, Boston, USA
Faculty:TNW: Science and Technology
Subject:44 medicine, 50 technical science in general, 54 computer science
Programme:Technical Medicine MSc (60033)
Link to this item:http://purl.utwente.nl/essays/76717
Export this item as:BibTeX
EndNote
HTML Citation
Reference Manager

 

Repository Staff Only: item control page