Deep Learning Approach for Non-Alcoholic Fatty Liver Recognition

Non-alcoholic fatty liver disease (NAFLD) is recognized as a burden on healthcare. In this study, we employed transfer learning with artificial intelligence to fine-tune a model using an ultrasound dataset for NAFLD, achieving a maximum accuracy of 100%. The research was designed as a binary classification task to develop a new framework for this dataset. The model architecture was reduced to only 1/59 of the original parameters while outperforming the conventional transfer learning methods in leave-one-out cross-validation (LOOCV) scenarios. The main contribution of this study is as follows: (1) establishing a novel workflow for ultrasound applications, (2) achieving a significant reduction in the number of parameters compared to traditional transfer learning methods, with a ratio of 1/59, and (3) demonstrating high accuracy rates using the new framework when fatty degeneration exceeds 25%.