Application of Deep Learning in Quantitative Analysis of 2‐Dimensional Ultrasound Imaging of Nonalcoholic Fatty Liver Disease

Abstract

To verify the value of deep learning in diagnosing nonalcoholic fatty liver disease (NAFLD) by comparing 3 image-processing techniques. A total of 240 participants were recruited and divided into 4 groups (normal, mild, moderate, and severe NAFLD groups), according to the definition and the ultrasound scoring system for NAFLD. Two-dimensional hepatic imaging was analyzed by the envelope signal, grayscale signal, and deep-learning index obtained by 3 image-processing techniques. The values of the 3 methods ranged from 0 to 65,535, 0 to 255, and 0 to 4, respectively. We compared the values among the 4 groups, draw receiver operating characteristic curves, and compared the area under the curve (AUC) values to identify the best image-processing technique. The envelope signal value, grayscale value, and deep-learning index had a significant difference between groups and increased with the severity of NAFLD (P < .05). The 3 methods showed good ability (AUC > 0.7) to identify NAFLD. Meanwhile, the deep-learning index showed the superior diagnostic ability in distinguishing moderate and severe NAFLD (AUC = 0.958). The envelope signal and grayscale values were vital parameters in the diagnosis of NAFLD. Furthermore, deep learning had the best sensitivity and specificity in assessing the severity of NAFLD.