Open Collections

Javadi, Golara

University of British Columbia

Ultrasound-guided needle biopsy with pathologic grading is the standard-of-care to guide the systematic biopsy of the prostate. Systematic transrectal ultrasound is blind to prostate pathology; and diagnostic accuracy is still one of the main clinical challenges in prostate cancer treatment and management. Hence, there is a clear need for improving US data and providing solutions to guide the biopsy procedure. Several methods including temporal enhanced UltraSound have been proposed to improve US-based tissue typing. The ultimate clinical goal is to display the cancer likelihood maps on B-mode ultrasound images in real time to help with indication of cancer in the core, decision support for biopsy targeting, and eventually reduc- ing the number of unnecessary biopsies. The objective of this dissertation is to deploy this by integration of temporal enhanced UltraSound and machine learning approaches. Towards fulfilling this objective, in this dissertation a weakly super- vised learning technique is utilized to learn from ultrasound image regions associ- ated with the corresponding data on patient level. To improve the prostate cancer detection we consider the nonstationarity nature of the data and use a complex neural network to find a better representation of the data in the embedding space for a better classification. Later, we automate reliable detection by estimating the model and label uncertainty. We finally show in order to improve the performance of prostate cancer classification, the label noise needs to be considered and in this work, the latter is done by implementing a label refinement technique.

Text

2022-10-19

Attribution-NonCommercial-NoDerivatives 4.0 International

http://hdl.handle.net/2429/82925

Electrical and Computer Engineering

University of British Columbia

UBCV

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