Fatty Liver Classification Via Risk Controlled Neural Networks Trained On Grouped Ultrasound Image Data

This article proposes a machine learning model for diagnosing fatty liver based on grouped ultrasound images. The model uses abstract representations of each image to deliver a subject-level diagnosis, eliminating the need for labeling each individual image. The model also incorporates a statistical method for controlling risk. Experiments were conducted with approval from the Taiwan Biobank and the Institutional Review Board at Academia Sinica.