Delineating regions of interest in MRI/S prostate scans for cancer diagnosis

Daniel Morgan


Supervised by Frank C Langbein; Moderated by Jing Wu

Prostate cancer is the second most common cancer in men worldwide. Earlier diagnosis combined with better staging and treatment has lead to a decrease in death rates for most developed countries. Magnetic resonance imaging and spectroscopy (MRI/S) plays an important role in differentiating malignant from benign prostate tissue. We are working on a project utilising machine learning techniques to devise an assistant to the radiologist to help in this differentiation.

The aim of this project is to devise a tool for the radiologist to delineate regions of interest, that potentially contain malignant tissue, in spatial MRI/S data / dicom images. It should make it simple for the radiologist to navigate through the data and mark the regions of interest and store the regions of interest (suitable to be used by our machine learning system as training and test data). You may further consider extending the tool to enable incorporating the regions of interest determined by an AI and enable adjustment/correction of these regions of interest by the radiologist to feed this back to the AI.

The tool may be implemented as a standard desktop or mobile (preferably Android) application. Potentially multiple versions of this project may be run for different platforms.

You must have excellent programming skills for this project with a focus on UIs and graphics/image processing, and relevant experience in the platform you wish to target. To be useful for our applications, we need any developed software or libraries to run under Linux and it must be released under a GNU GPL v3 or a compatible license.

Initial Plan (04/02/2019) [Zip Archive]

Final Report (10/05/2019) [Zip Archive]

Publication Form