Deep Learning Elastography for Assessing Liver Fibrosis
- VernacularTitle:深度学习分析剪切波弹性图像评估肝纤维化
- Author:
Wen-bo CHEN
1
;
Xue LU
2
;
Jie-yang JIN
2
;
Rong-qin ZHENG
2
Author Information
1. Department of Ultrasound， Qingyuan People's Hospital， Qingyuan 511500， China
2. Department of Ultrasound， The Third Affiliated Hospital of Sun Yat-sen University， Guangzhou 510630， China
- Publication Type:Journal Article
- Keywords:
deep learning;
shear wave elastography;
liver fibrosis
- From:
Journal of Sun Yat-sen University(Medical Sciences)
2021;42(2):294-301
- CountryChina
- Language:Chinese
-
Abstract:
ObjectiveTo assess the diagnostic performance of deep learning elastography （DLE） for liver fibrosis. MethodsTotally 545 chronic liver disease patients with liver biopsy were retrospectively enrolled. The results of DLE， two dimensional shear wave elastography （2D-SWE）， serological markers and transient elastography （TE） were collected. Area under receiver operating curve （AUC） was calculated and inter-compared while evaluating liver fibrosis. Then we tested the diagnostic performance of the trained DLE model in different validation groups when evaluating the same liver fibrosis stage， respectively， to assess the stability of DLE. ResultsDLE showed statistically significantly （P＜0.05） better results than any other methods. When evaluating F=4， F≥3 and F≥2， the AUC of DLE was 0.99， 0.98 and 0.92， respectively. 2D-SWE showed a second high diagnostic performance， while the AUCs were 0.89， 0.86 and 0.86， respectively. Little difference in diagnostic performance was showed among other methods， while the highest AUC was no more than 0.81. Besides， no statistical difference was showed among the three validation groups， while accessing the same liver fibrosis stage. ConclusionsDLE can be used to accurately assess liver fibrosis， whose diagnostic performance is higher than that of 2D-SWE， serological markers and TE. Moreover， with good stability， DLE is expected to become a new method for noninvasive assessment of liver fibrosis.
