Principal Investigator: Mr Brett Kuprel
Department: Stanford University
Stanford, USATags: 50783, deep learning, fatty liver, preventive-health
Livers have the second longest transplant waitlist in the US next to kidneys. Alcohol and obesity are known causes of fat build up in the liver. This fat build up leads to scarring. 1 in 4 people in the US and Europe have livers that are over 5% fat and experience no symptoms. The only good options for quantifying liver fat percentage are an invasive liver biopsy or an expensive MRI, both of which require significant suspicion of liver damage or other problems to justify. A tighter feedback loop would allow healthy minded individuals to correct deleterious behavior before reversible liver fat becomes irreversible liver scarring.
We aim to combine the latest advances in AI with routine blood and urine measurements to create a tool with which the healthy man and woman can better monitor their liver health. The current medically accepted fatty liver calculator uses 2 blood measurements in conjunction with body weight and height. It was trained on data collected from 496 people. The UK Biobank dataset contains data from over half a million people, and tons of blood and urine measurements. Our scientific rationale is that we can use this enormous dataset to train a more reliable fatty liver estimator.
When granted access to the dataset, we expect the project to be completed in under 3 months. If the trained estimator is reliable, we plan to release it to the public as an app. The public health impact of this work will be to allow better monitoring of liver health with data collected from routine blood and urine lab measurements.