University of Twente Student Theses


Bayesian uncertainty estimation of deep learning carotid artery vessel wall segmentation

Buurman, Sophie (2022) Bayesian uncertainty estimation of deep learning carotid artery vessel wall segmentation.

[img] PDF
Abstract:Monitoring patients with atherosclerosis demands measurements of the thickness of the carotid artery vessel wall. An accurate segmentation is essential for these measurements, however manual acquisition is extremely time consuming. Recently, Alblas et al. [2] proposed a fully automatic method for vessel wall segmentation on 3D MRI images, ensuring ring-shaped segmentations. The method returns contour points describing two nested circles, representing the lumen and outer wall on each axial slice of the image. Although very successful, the model returns a prediction regardless of the underlying image quality. This can be problematic in medical images that contain regions of noise or artefacts, as the model should indicate the segmentations are uncertain around those regions. Therefore, we propose the use of dropout layers in the convolutional neural network of Alblas et al., introducing stochasticity in the network. These dropout layers can be used to approximate the posterior predictive distribution by passing multiple stochastic inferences through the network. The predictive mean and variance are calculated for each of the predictive contour points. As we hypothesized, we observe a substantial higher variance for low quality image data, as well as near the carotid bifurcation.
Item Type:Essay (Bachelor)
Faculty:EEMCS: Electrical Engineering, Mathematics and Computer Science
Subject:31 mathematics, 44 medicine
Programme:Applied Mathematics BSc (56965)
Link to this item:
Export this item as:BibTeX
HTML Citation
Reference Manager


Repository Staff Only: item control page