AI Case Study
Researchers at University of Washington explore diagnosis of early onset of pancreatic cancer by identifying increased bilirubin levels in sclera from selfies with 89% accuracy
Researchers are developing an App,BiliScreen, that can test for increased Bilirubin levels from pictures of sclera to detect jaundice which is often the first symptom of pancreatic cancer. Pancreatic cancer is non-symptomatic until its too late and this could be a non-invasive way of diagnosing the disease early enough to treat it. So far the algorithm has achieved 89% accuracy.
Public And Social Sector
Education And Academia
"University of Washington researchers are developing an app that could allow people to easily screen for pancreatic cancer and other diseases — by snapping a smartphone selfie.
BiliScreen uses a smartphone camera, computer vision algorithms and machine learning tools to detect increased bilirubin levels in a person’s sclera, or the white part of the eye. The app is described in a paper to be presented Sept. 13 at Ubicomp 2017, the Association for Computing Machinery’s International Joint Conference on Pervasive and Ubiquitous Computing.
Elevated bilirubin levels is one of earliest symptoms of pancreatic cancer and can help in early detection.
Looking beyond jaundice, quantitative visual examination of the sclera can yield other fruitful observations. Osteogenesis imperfecta, a genetic disorder that results in brittle bones, produces a blue tinge in the sclera. Diabetes results in fewer capillaries, dilated macrovessels, and changes in the curvature in the covering of the sclera. Hyperemia and conjunctivis can affect both the amount and contrast of the blood vessels on the scleral surface. BiliScreen’s sclera segmentation algorithm could be used as a starting point for a system that searches for these symptoms and others."
R And D
Pancreatic cancer has one of the highest fatality rates — with a five-year survival rate of 9 percent — partly because external symptoms are shown too late
In first round of testing, correctly identified cases of concern 89.7 percent of the time, compared to the blood test currently used, with a sensitivity of 92.8% and a specificity of 94.3%..
"The models use random forest regression and are trained through 10-fold cross-validation across participants"