Professor Ghassan Hamarneh's computerized image analysis and machine-learning algorithms will help doctors detect gastrointestinal cancer more quickly and less invasively than current methods.

Research collaboration set to speed diagnosis of gastrointestinal cancer

April 25, 2017

A team of scientists, including Simon Fraser University (SFU) computing science professor Ghassan Hamarneh, is poised to create a new screening tool that could help clinicians identify early signs of gastrointestinal cancer.

The three-year interdisciplinary project recently received funding from a National Sciences and Engineering Research Council (NSERC) Collaborative Health Research Project (CHRP) grant of more than $700,000.

Gastrointestical cancer is the second leading cause of cancer death worldwide, second only to lung cancer. It also has the lowest survival rates of any cancer, at just 30 per cent.

“Early detection is crucial for the success of gastrointestinal cancer therapy, yet less than half of new gastrointestinal cancer cases in Canada are identified at the most curable stage,” says Hamarneh.

Doctors typically use a white light endoscope—a flexible tube with a light and camera attached—to find and diagnose gastrointestinal cancer by examining the surface tissue of the digestive tract.

With early gastrointestinal cancer lesions, however, the changes in shape or colour are subtle and can be difficult to detect during a standard endoscopy procedure. Patients may need to undergo multiple invasive biopsy and screening procedures to receive a definitive diagnosis. During this time, the cancer may further develop or spread.  

To tackle this problem, Hamarneh will collaborate with researchers from the University of British Columbia (UBC), the Vancouver General Hospital (VGH) and the BC Cancer Agency to design and build a new type of endoscope and imaging system over the next three years.

The innovative system will provide clinicians with rapid three-dimensional imaging of the entire gastrointestinal wall that could both speed diagnosis and reduce invasive biopsies.

Hamarneh and his team in SFU’s Medical Image Analysis lab will develop computerized image analysis and machine-learning algorithms to enhance and analyze the images.

Ultimately, the researchers hope to create a system that can automatically detect the probability of cancer versus a less serious gastric condition. This could help clinicians make decisions regarding treatment or surgery based on objective, quantitative insights.

“Every time there is an improvement in imaging, we expect it to translate into an improvement in diagnosis,” says Hamarneh.  

Hararneh will partner with UBC professors Miu Chiao, Ken Chou and Ryozo Nagamune, VGH clinicians Isabella Tai and Baljinder Salh, and BC Cancer Agency scientist Haishan Zeng.

How it works

The researchers will tap into an emerging technology called confocal endoscopy, which uses lasers to produce three-dimensional (3D) cross-sectional images of the gastrointestinal surface and subsurface tissue in real-time.

In addition, it is capable of visualizing cell and tissue features at one-millionth of a metre—the size of a single red blood cell.

But there’s the rub.

Due to its small size, good quality images cannot always be obtained. So Hamarneh must develop sophisticated algorithms capable of looking for subtle patterns in lower-quality images.

Hamarneh and his team come with experience in this area. They have previously developed biomedical-imaging systems that use machine-learning techniques to identify signs of ovarian, skin and kidney cancer.

“We start with hundreds or thousands of images of cancerous and benign lesions that we feed into the computer as training data,” says Hamarneh.

“The goal is that, from seeing so many previous examples, the computer will learn how to predict what is cancerous and what is benign.”

Using mathematical optimization, the computer can be taught to “hunt” for specific clues and patterns of gastrointestinal cancer, such as colour, texture and shape.

“Eventually, we also want to take this into the realm of deep learning,,” says Hamarneh. “We want to train our machine to discover indications of cancer by itself, by analyzing complex combinations and transformations of pixel colours within an image.”

For the patient, this can mean a better chance of definitive and early therapy.

“If you think about the patient, it all falls into place,” says Hamarneh, referring to his source of motivation. “The patient’s health is paramount. We need to diagnose problems early on using a test or procedure that is widely available, cheap, quick and accurate, to ensure that this life-saving screening will take place.”