Artificial intelligence for liver disease detection using ultrasound images
Artificial intelligence (AI) could facilitate an automated and accurate early detection of liver diseases. In this study, we optimized via transfer learning Visual Geometry Group 19 (VGG19), a convolutional neural network (CNN) architecture to classify liver ultrasound images into different disease classes. The dataset used to develop the classifier model consists of 1895 ultrasound images generated from a clinical study of Filipino cohorts. Results show that the fine-tuned classifier model achieved an overall testing accuracy of 86.7 ± 1.1% with both cirrhotic and hepatic classes achieving an AUC score of 1.00, while the fatty, normal, and parenchymal classes achieved AUC scores of 0.96, 0.95, and 0.96, respectively. To a certain extent, this study illustrates the applicability of an AI-driven model for an automated disease detection using medical images.