Logo image
Convolutional Neural Network Classification of Liver Fibrosis Stages Using Ultrasonic Images Colorized by Features of Echo-Envelope Statistics
會議論文集

Convolutional Neural Network Classification of Liver Fibrosis Stages Using Ultrasonic Images Colorized by Features of Echo-Envelope Statistics

Akiho Isshiki, Dar-In Tai, Po-Hsiang Tsui, Kenji Yoshida, Tadashi YamaguchiShinnosuke Hirata
MEDICAL IMAGING AND COMPUTER-AIDED DIAGNOSIS, MICAD 2022, 卷.810, 頁碼.441-451
Lecture Notes in Electrical Engineering
01/01/2023
Web of Science ID: WOS:001491661400036

摘要

Computer Science Computer Science, Artificial Intelligence Computer Science, Theory & Methods Engineering Engineering, Biomedical Life Sciences & Biomedicine Radiology, Nuclear Medicine & Medical Imaging Science & Technology Technology
The progression of liver fibrosis is the most important indicator that determines the prognosis of patients with diffuse liver disease. Variations in tissue structure triggered by liver fibrosis severely affect the texture and contrast of the ultrasound image. Therefore, progression can be non-invasively evaluated by analyzing ultrasound images. The convolutional neural network (CNN) classification of liver fibrosis stages using ultrasound images has also been studied. In previous studies, grayscale ultrasound images obtained using conventional ultrasound scanners were adopted as the input images. In this study, the modulation and colorization of the ultrasound images by the echo-envelope statistics that correspond to the texture and contrast of the ultrasound images have been proposed. In the proposed method, the colorized ultrasound image in RGB representation comprises the original image and two images modulated by different features of the echo-envelope statistics. Accordingly, the effect enhancement of tissue-structure variation by the colorization of the ultrasound images is promising in improving the accuracy of CNN classification. Therefore, CNN classification of the ultrasound images colorized by their 1st- and 3rd-order moments is demonstrated via the transfer learning of the VGG-16 pretrained network.

相關連結

指標

1 檢視次數

詳細資料

Logo image