Why it matters to you
The University of Washington’s new mobile app can screen for one of the deadliest forms of cancer in a non-invasive way.
There have been a few examples of research we covered that have used machine learning technology to diagnose cancer. None of them have been quite like the University of Washington’s BiliScreen project, however. In an attempt to diagnose cancer, researchers at the university have developed a smartphone app that is capable of carrying out a screening by asking users to snap a quick selfie. Could there be a more 2017 medical diagnosis tool?
But if the idea sounds whimsical, the problem it is helping to solve is anything but. The type of cancer the researchers are interested in is pancreatic cancer, a particularly nasty type of cancer with a five-year survival rate of just nine percent. (Pancreatic cancer was the cause of death of former Apple CEO and co-founder Steve Jobs). Making pancreatic cancer hard to spot is the fact that it does not display obvious telltale symptoms which allow people to catch it prior to it spreading. That is where the University of Washington’s new smartphone app enters the frame.
“BiliScreen is a smartphone app we are developing to quantify the extent of jaundice in an individual,” Alex Mariakakis, a doctoral student at the Paul G. Allen School of Computer Science and Engineering, told Digital Trends. “Jaundice is the yellowing of the skin and eyes due to the buildup of a compound called bilirubin in the blood. Jaundice becomes obvious to the naked eye at 3.0 mg/dl, but clinicians start to be concerned at only 1.3 mg/dl, leaving a gap where detection could be important. There are many reasons why someone may become jaundiced, including hepatitis and Gilbert’s syndrome. What we are most excited about, [though], is the fact that jaundice is one of earlier symptoms that appears in people who have pancreatic cancer.”
The app works by using a smartphone camera, along with computer vision algorithms and machine-learning tools, to look for increased bilirubin levels in the white part of a person’s eye. In an initial clinical study of 70 people, the app — along with a 3D-printed box that controls the eye’s exposure to light — was able to correctly identify cases of concern 89.7 percent of the time.
As Mariakakis notes, jaundice does not necessarily equate to pancreatic cancer, but recognizing elevated levels of it in the eye could be a sign that an individual should consult a physician.
As exciting as this is, Mariakakis points out it will take longer before it is ready to be made available as a clinical tool. “Giving people information that could lead them to believe they have a serious medical condition is not something we take lightly, so there is more research to be done before we get to that point,” he said.
A paper describing the project can be read here. It will be presented in September at Ubicomp 2017, the Association for Computing Machinery’s International Joint Conference on Pervasive and Ubiquitous Computing.