Non-Alcoholic Fatty Liver Disease (NAFLD) is a highly prevalent disease affecting 25% of adults.
It often goes undetected until significant, irreversible damage has occurred. However, if the disease is detected early in its progression, patients can make lifestyle changes to prevent damage from occuring.
Today, determining the presence and severity of NAFLD requires an invasive and expensive liver biopsy procedure.
Our team leveraged medical professionals to accurately label organs, veins, and masses in abdominal ultrasound images, which were subsequently used to train state of the art object detection, segmentation, and classification
algorithms. Based on our research, we are able to correctly classify livers as being positive or negative for NAFLD with 94% accuracy.
The first stage of our pipeline detects the presence of the kidney and liver, as well as various structures within the liver (veins and lesions) in the input ultrasound image. If the liver and kidney are both detected in an
image, we pass the image to a second stage model to outline the liver and the kidney cortex and medulla. These key organs and structures are extracted from the original image and passed to our third and final model, which classifies the presence
and degree of NAFLD.
