Use this URL to cite or link to this record in EThOS:
Title: Automatic analysis of medical images for change detection in prostate cancer
Author: Ghavami, Nooshin
ISNI:       0000 0004 9353 529X
Awarding Body: UCL (University College London)
Current Institution: University College London (University of London)
Date of Award: 2020
Availability of Full Text:
Access from EThOS:
Full text unavailable from EThOS. Please try the link below.
Access from Institution:
Prostate cancer is the most common cancer and second most common cause of cancer death in men in the UK. However, the patient risk from the cancer can vary considerably, and the widespread use of prostate-specific antigen (PSA) screening has led to over-diagnosis and over-treatment of low-grade tumours. It is therefore important to be able to differentiate high-grade prostate cancer from the slowly- growing, low-grade cancer. Many of these men with low-grade cancer are placed on active surveillance (AS), which involves constant monitoring and intervention for risk reclassification, relying increasingly on magnetic resonance imaging (MRI) to detect disease progression, in addition to TRUS-guided biopsies which are the routine clinical standard method to use. This results in a need for new tools to process these images. For this purpose, it is important to have a good TRUS-MR registration so corresponding anatomy can be located accurately between the two. Automatic segmentation of the prostate gland on both modalities reduces some of the challenges of the registration, such as patient motion, tissue deformation, and the time of the procedure. This thesis focuses on the use of deep learning methods, specifically convolutional neural networks (CNNs), for prostate cancer management. Chapters 4 and 5 investigated the use of CNNs for both TRUS and MRI prostate gland segmentation, and reported high segmentation accuracies for both, Dice Score Coefficients (DSC) of 0.89 for TRUS segmentations and DSCs between 0.84-0.89 for MRI prostate gland segmentation using a range of networks. Chapter 5 also investigated the impact of these segmentation scores on more clinically relevant measures, such as MRI-TRUS registration errors and volume measures, showing that a statistically significant difference in DSCs did not lead to a statistically significant difference in the clinical measures using these segmentations. The potential of these algorithms in commercial and clinical systems are summarised and the use of the MRI prostate gland segmentation in the application of radiological prostate cancer progression prediction for AS patients are investigated and discussed in Chapter 8, which shows statistically significant improvements in accuracy when using spatial priors in the form of prostate segmentations (0.63 ± 0.16 vs. 0.82 ± 0.18 when comparing whole prostate MRI vs. only prostate gland region, respectively).
Supervisor: Not available Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID:  DOI: Not available