AUTHOR=Hirata Shinnosuke , Isshiki Akiho , Tai Dar-In , Tsui Po-Hsiang , Yoshida Kenji , Yamaguchi Tadashi 

TITLE=Convolutional neural network classification of ultrasound images by liver fibrosis stages based on echo-envelope statistics

JOURNAL=Frontiers in Physics

VOLUME=Volume 11 - 2023

YEAR=2023

URL=https://www.frontiersin.org/journals/physics/articles/10.3389/fphy.2023.1164622

DOI=10.3389/fphy.2023.1164622

ISSN=2296-424X

ABSTRACT=In the diagnosis and the follow-up for patients with diffuse liver disease, it is most important to assess what stage liver fibrosis is on. The tissue structure in the fibrotic liver is reflected to the texture and contrast of the ultrasound image whose pixel brightness indicate the intensities of echo envelopes. Therefore, the progression of liver fibrosis can be non-invasively evaluated by analyzing the ultrasound image. The convolutional-neural-network (CNN) classification of ultrasound images is applied for the liver fibrosis estimation. In this study, the colorization of the ultrasound images using echo-envelope statistics that correspond to features of the images have been proposed for improving the accuracy of the CNN classification. In the proposed method, the ultrasound image is modulated by the 3rd- and 4th-order moments of pixel brightness. Then, the two modulated images and the original image are synthesized to the color image in RGB representation. The colorized ultrasound images were classified via the transfer learning of the VGG-16 to evaluate the effect of the colorization. In 80 ultrasound images with the liver fibrosis stages from F1 to F4, 38 images were accurately classified by the CNN using the original ultrasound images, whereas 47 images could be classified by the proposed.