Automatic Diagnosis of Significant Liver Fibrosis from Ultrasound B-mode Images Using a Handcrafted-Feature-Assisted Deep Convolutional Neural Network

• Published in: IEEE journal of biomedical and health informatics Vol. PP; no. 10; pp. 1 - 12
• Main Authors: Liu, Zhong, Huang, Bin, Wen, Huiying, Lu, Zhicheng, Huang, Qicai, Jiang, Meiqin, Dong, Changfeng, Liu, Yingxia, Chen, Xin, Lin, Haoming
• Format: Journal Article
• Language: English
• Published: United States IEEE 01.10.2023; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: Antiviral agents; Artificial neural networks; Attention; Biological system modeling; Biopsy; Deep learning; Diagnosis; Feature extraction; feature fusion; Fibrosis; Image processing; Image segmentation; Liver; Liver diseases; Machine learning; Medical diagnosis; Medical imaging; Model accuracy; Neural networks; significant liver fibrosis; Ultrasonic imaging; Ultrasound
• ISSN: 2168-2194; 2168-2208
• DOI: 10.1109/JBHI.2023.3295078

The accurate diagnosis of significant liver fibrosis (<inline-formula><tex-math notation="LaTeX">\rm \geq</tex-math></inline-formula>F2) in patients with chronic liver disease (CLD) is critical, as <inline-formula><tex-math notation="LaTeX">\pmb \mathrm{\geq }</tex-math></inline-formula>F2 is a crucial factor that should be considered in selecting an antiviral therapy for these patients. This paper proposes a handcrafted-feature-assisted deep convolutional neural network (HFA-DCNN) that helps pathologists automatically and accurately diagnose significant liver fibrosis from ultrasound (US) brightness (B)-mode images. The HFA-DCNN model has three main branches: one for automatic region of interest (ROI) segmentation in the US images, another for attention deep feature learning from the segmented ROI, and the third for handcrafted feature extraction. The attention deep learning features and handcrafted features are fused in the back end of the model to enable more accurate diagnosis of significant liver fibrosis. The usefulness and effectiveness of the proposed model were validated on a dataset built upon 321 CLD patients with liver fibrosis stages confirmed by pathological evaluations. In a fivefold cross validation (FFCV), the proposed model achieves accuracy, sensitivity, specificity, and area under the receiver-operating-characteristic (ROC) curve (AUC) values of 0.863 (95% confidence interval (CI) 0.820-0.899), 0.879 (95% CI 0.823-0.920), 0.872 (95% CI 0.800-0.925), and 0.925 (95% CI 0.891-0.952), which are significantly better than those obtained by the comparative methods. Given its excellent performance, the proposed HFA-DCNN model can serve as a promising tool for the noninvasive and accurate diagnosis of significant liver fibrosis in CLD patients.