As five new PhD students join our team of scientists and clinicians fighting prostate cancer, we catch up with one student coming to the end of his study. Our guest blogger this week is Jonathan Francis Roscoe, a PhD student in the Vision, Graphics and Visualisation group at Aberystwyth University, supervised by Hannah Dee and Reyer Zwiggelaar. Back in 2011 we awarded the team a grant, funded by the Hoover Foundation, to fund Jonathan’s project.

24 Sep 2014
In - Blog Research

Jonathan Francis Roscoe, PhD Student

It was extremely exciting to discover we had been awarded a grant from Prostate Cancer UK. It’s been a busy three years, but the project is coming to an end so it’s a good time to let you know what we’ve been up to.

As many of you will know, prostate biopsy (where small samples of tissue are removed with a needle) is the definitive method for finding out if there is cancer in a man’s prostate. Unfortunately, there are a number of issues with this method. The majority of biopsies are performed using ultrasound to look at the prostate. While this is cheap and readily available it provides only a basic image of the prostate making it difficult to target areas where there might be cancer.

Magnetic resonance imaging (MRI) is a much clearer means of looking at the prostate, but is costly and time consuming. And while biopsy can be performed under MRI, it requires specialist equipment and isn’t currently standard practice.

The aim of our project was to take standard MRIs and combine them with ultrasound in order to improve the accuracy of the ultrasound-guided biopsy. You could think of it as a map to highlight “high risk” regions, so that the urologist has an easier time identifying suspicious areas.

One of the first steps towards this was to build a 3D computer model of the prostate that provided a typical map of tissue types, including locations most likely to contain cancer. An MRI scan works by taking images at several locations, producing a number of images representing cross-sections of the body – known as slices. A typical scan is made of around 20 slices, which when stacked together provide a clear image of the prostate. To construct our 3D model we combined over 1,200 MRI slices from more than 50 men with prostate cancer. Each slice had been annotated by radiologists to indicate three major regions of interest – the prostate capsule, the central region and areas of cancer (if there were any).

We applied image processing techniques found in hand-writing recognition and face morphing tools (which you may have seen popularised recently as mobile apps). This allowed us to combine all of the images into a single model, identifying areas where tumours commonly appear throughout the prostate. Although no two cases are alike, combining a variety allows us to find common characteristics.

We had a lot of problems to tackle and we’ve done a lot of research into potential techniques. Much of my research has been inspired by mammographic and other medical image analysis techniques that have similar characteristics. (A mammogram is an image of a woman’s breast used to try and spot cancer.) I am continually looking for methods that might be applicable to the MRI and ultrasound images of the prostate.

At the moment we’re wrapping up investigation into a novel technique for detecting areas of cancer in MRI that works by combining different characteristics of the image to differentiate structures. Every image, whether it’s from a camera or an MRI scan has a number of features that can tell us about the content. For example, in a landscape if you were searching for sky, you might look for blue regions. If you want to find a road you might look for straight lines.

Often, cancer areas are much darker though this alone isn’t a good enough test. In our method we use two major characteristics. The first is intensity (how bright or dark a spot on the image is) and the second is microstructure (we count small straight lines, corners, etc. at a scale a human might not see). Together, these give us a new way of describing sections of an image.

Once we’ve done this, we can combine it with our earlier work in 3D modelling to project it onto ultrasound images (which don’t have a lot of useful information) to improve their reliability.

Over the next six months I will be reviewing the work I’ve done so far, and putting it together with similar research published by authors from around the world in order to hone our methodology and form my thesis. I don’t yet know what I will be doing afterwards, but I am planning to investigate further research work. I have thoroughly enjoyed the opportunity Prostate Cancer UK gave me. I hope my work will not only be of value to others working in medical image analysis but also to men with, or at risk of prostate cancer. In particular, this work should be a step towards better diagnosis and staging.

I also want to say good luck to any new research students starting soon. It’s a long journey with lots of different paths to take. My recommendation is to collaborate with peers in your own department as often as you can. There are lots of great ideas you can get by interacting with people in other disciplines.

Doing my PhD has given me the opportunity to implement my own ideas and interact with people from around the world. I’ve attended a variety of events and conferences and am thrilled to see the number and variety of people investigating the diagnosis and treatment of prostate cancer. It’s great to have the opportunity to discuss ideas.

It’s been really encouraging to see the boom in awareness of prostate cancer in the UK over the past three years, and I’ve been proud to be a part of the research. I look forward to keeping up with the work of future PhD students in Progress magazine.

comments powered by Disqus