The Pancreatic Cancer Action Network reported good news last year: the 5-year survival rate for patients diagnosed with the disease increased from 6% to 9%. Because pancreatic cancer is asymptomatic in early stages, most cases metastasize before diagnosis. In the same report, the group estimated an annual 43,090 American deaths from the disease and 53,670 newly diagnosed cases.

University of Washington researchers developed BiliScreen to assist with early pancreatic cancer diagnosis. BiliScreen is a smartphone app that detects bilirubin levels not visible to the human eye. Bilirubin buildup in the blood causes jaundice, a yellowish discoloration of the skin and eyes. Jaundice can also indicate bile duct obstruction, liver disease, or excessive red blood cell breakdown. In most people, the sclera, the white part of the eye, is more sensitive to bilirubin buildup than skin, but it’s still hard to detect changes. BiliScreen uses machine learning algorithms and smartphone camera images of test subjects’ eyes. In a relatively small sample of 70 people, the app correctly identified significant bilirubin levels 89.7% of the time. The UW researchers tested two additional tools in the study, a light-controlling 3D-printed box and paper glasses with color blocks to use as references. Test results using the box were slightly better than with the glasses.

According to the UW researchers, the next steps include larger subject sample sizes and improved app and smartphone usability. This test, like research by Google scientists who entailed big data and deep learning with retinal images to predict heart disease and measure blood pressure, indicates the potential of much stronger predictive and diagnostic tools when imagery and artificial intelligence work in conjunction.