COMPUTERIZED IDENTIFICATION OF THE PHASES OF LIVER FIBROSIS BY ULTRASONOGRAPHY: QUANTITATIVE STUDY OF DEEP CONVOLUTIONAL NEURAL NETWORK

Abstract

This study aimed to evaluate the effectiveness of a widely used and well-respected deep convolutional neural network and to determine the computerized detection of liver fibrosis episodes using ultrasonography. In three months of 2024, a quantitative study was conducted in several settings in Bangladesh. The design of the study was quantitative; therefore informed consent was not required. While non-invasive techniques like transient elastography are frequently employed to assess liver fibrosis, mistakes may arise in situations involving a constricted liver or ascites. In this work, we used US photos to fine-tune the model following transfer learning on ImageNet. In this study, the learning-rate scheduler and optimization algorithm was Adam optimizer, respectively, while the loss function used for training the models was CrossEntropyLoss. Accuracy, sensitivity, specificity, and positive and negative likelihood ratios were used to assess DCNN performance. Furthermore, the effectiveness of DCNNs was evaluated using the area under the receiver operating characteristic curve (AUC) with a 95% confidential interval for five-level categorization. 49 were the median age (IQR: 42–58). F0 is generally easy to get, and 33.2% of the dataset consists of such data. 33.2% of the dataset was made up of liver fibrosis stage 4, or F4. However, because so few individuals were evaluated in the early stages of hepatic fibrosis, the proportions of F1 (13.2%), F2 (8.4%), and F3 (24.2%) were quite low. Using a computer vision method, a data augmentation of the images expands the size of a limited dataset. In the final analysis, researchers have shown that DCNNs can accurately classify METAVIR grade employing traditional US pictures. The DCNNs-based diagnosis of liver fibrosis using B-mode images will be an effective instrument for promoting radiologists in the clinical setting, as US imaging is a commonly accessible method and is primarily employed in periodic subsequent studies of patients with persistent liver disease. Additional enhancement and verification might be necessary, though.