How GPU-Powered Cellular Analysis Could Help Avoid Unnecessary Prostate Cancer Biopsies

Some 1.3 million men in North America each year undergo biopsies to determine whether they have prostate cancer. While these are ordered only after multiple tests indicate the possible presence of cancer, many are ultimately proven to have been unnecessary.

Researchers at the University of Alberta hope to change that with a new testing method that relies on GPU computing — and which could prevent up to half of those biopsies from happening.

The de facto screening method, known as a Prostate-Specific Antigen (PSA) test, is “not very reliable, and lots of men undergo biopsies for little benefit, and they risk infection and other side effects in the process,” said John Lewis, the Frank and Carla Sojonky Chair of Prostate Cancer Research funded by the Alberta Cancer Foundation.

Focusing on Vesicles

Lewis’s team has taken a different approach it’s calling the Extracellular Vesicle Fingerprint Predictive Score. The EV-FPS involves analyzing biomarker data from extracellular vesicles, fluid-filled sacs that enable communication between cells. By scrutinizing vesicles, researchers can predict the presence of cancer cells with more precision than with PSA tests.

The challenge is that the new test analyzes up to 5 million vesicles using four biomarkers each, resulting in a total of 20 million data points per patient. But lurking in these vesicles are those that began in the prostate gland, each serving as a sort of map of the cells from which they originated.

That’s where NVIDIA GPUs and deep learning tools come into play.

“With that much information, there was simply no way to identify the key patterns in the data by eye, which led us to GPUs,” said Lewis.

To boost its analysis capabilities, the team paired an NVIDIA GeForce GTX GPU with Mathworks’ MATLAB numerical computing software, as well as CUDA version 7.5 and cuDNN version 5 for deep learning. In doing so, the researchers were able to train a convolutional neural network to perform image-based analysis of all of that data.

Lewis’s team used an approach known as fivefold cross validation, in which patient data was divided into five groups, with one held out of each round of training. This improves the accuracy of the model by ensuring that every patient’s data is used once in the hold-out group during evaluation.

Improved Accuracy = Fewer Biopsies

The results speak for themselves: Lewis said that in validating the EV-FPS in 410 patients, his team has boosted the accuracy of cancer detection by 40 percent compared with PSA tests. That could translate to as many as half of those 1.3 million men avoiding a biopsy.

That kind of impact led the team to commercialize the new test, and the team has spun off a company, called Nanostics, to do just that. The EV-FPS has been packaged as a product called Clarity DX. Lewis, who serves as CEO of Nanostics, said it will hit the market in mid-2018 as a screening test. The company plans to seek FDA approval of the product so that it can make specific claims about its performance.

And this may be just the start. Lewis foresees this new ability to analyze extracellular vesicle biomarkers as having a potential impact on screenings of other cancers or neurodegenerative diseases.

Feature image: University of Alberta prostate cancer researcher Dr. John Lewis, left, works with graduate student Srijan Raha. Credit: University of Alberta

The post How GPU-Powered Cellular Analysis Could Help Avoid Unnecessary Prostate Cancer Biopsies appeared first on The Official NVIDIA Blog.