Open Collections

Machine learning for MRI-guided prostate cancer diagnosis and interventions

University of British Columbia

Prostate cancer is the second most prevalent cancer in men worldwide. Magnetic Resonance Imaging (MRI) is widely used for prostate cancer diagnosis and guiding biopsy procedures due to its ability in providing superior contrast between cancer and adjacent soft tissue. Appropriate clinical management of prostate cancer critically depends on meticulous detection and characterization of the disease and precise biopsy procedures if necessary. The goal of this thesis is to develop computational methods to aid radiologists in diagnosing prostate cancer in MRI and planning necessary interventions. To this end, we have developed novel methods for assessing probability of clinically significant prostate cancer in MRI, localizing biopsy needles in MRI, and providing segmentation of structures such as the prostate gland. The proposed methods in this thesis are based on supervised machine learning techniques, in particular deep convolutional neural networks (CNNs). We have also developed methodology that is necessary in order for such deep networks to eventually be useful in clinical decision-making workflows; this spans the areas of domain adaptation, confidence calibration, and uncertainty estimation for CNNs. We used domain adaptation to transfer the knowledge of lesion segmentation learned from MRI images obtained using one set of acquisition parameters to another. We also studied predictive uncertainty in the context of medical image segmentation to provide model confidence (i.e expectation of success) at inference time. We further proposed parameter ensembling by perturbation for calibration of neural networks.

Text

2020-10-22

Attribution-NonCommercial-NoDerivatives 4.0 International

http://hdl.handle.net/2429/76329

Electrical and Computer Engineering

University of British Columbia

UBCV

http://creativecommons.org/licenses/by-nc-nd/4.0/