Electronic Thesis and Dissertation Repository

Degree

Doctor of Philosophy

Program

Biomedical Engineering

Supervisor

Aaron Fenster

2nd Supervisor

Aaron D. Ward

Joint Supervisor

Abstract

Segmentation of the prostate in medical images is useful for prostate cancer diagnosis and therapy guidance. However, manual segmentation of the prostate is laborious and time-consuming, with inter-observer variability. The focus of this thesis was on accuracy, reproducibility and procedure time measurement for prostate segmentation on T2-weighted endorectal magnetic resonance imaging, and assessment of the potential of a computer-assisted segmentation technique to be translated to clinical practice for prostate cancer management. We collected an image data set from prostate cancer patients with manually-delineated prostate borders by one observer on all the images and by two other observers on a subset of images. We used a complementary set of error metrics to measure the different types of observed segmentation errors. We compared expert manual segmentation as well as semi-automatic and automatic segmentation approaches before and after manual editing by expert physicians. We recorded the time needed for user interaction to initialize the semi-automatic algorithm, algorithm execution, and manual editing as necessary. Comparing to manual segmentation, the measured errors for the algorithms compared favourably with observed differences between manual segmentations. The measured average editing times for the computer-assisted segmentation were lower than fully manual segmentation time, and the algorithms reduced the inter-observer variability as compared to manual segmentation. The accuracy of the computer-assisted approaches was near to or within the range of observed variability in manual segmentation. The recorded procedure time for prostate segmentation was reduced using computer-assisted segmentation followed by manual editing, compared to the time required for fully manual segmentation.

Download

Share

COinS