Ultrasonic Assessment of Liver Fibrosis Using One-Dimensional Convolutional Neural Networks Based on Frequency Spectra of Radiofrequency Signals with Deep Learning Segmentation of Liver Regions in B-Mode Images: A Feasibility Study

Abstract

The early detection of liver fibrosis is of significant importance. Deep learning analysis of ultrasound backscattered radiofrequency (RF) signals is emerging for tissue characterization as the RF signals carry abundant information related to tissue microstructures. However, the existing methods only used the time-domain information of the RF signals for liver fibrosis assessment, and the liver region of interest (ROI) is outlined manually. In this study, we proposed an approach for liver fibrosis assessment using deep learning models on ultrasound RF signals. The proposed method consisted of two-dimensional (2D) convolutional neural networks (CNNs) for automatic liver ROI segmentation from reconstructed B-mode ultrasound images and one-dimensional (1D) CNNs for liver fibrosis stage classification based on the frequency spectra (amplitude, phase, and power) of the segmented ROI signals. The Fourier transform was used to obtain the three kinds of frequency spectra. Two classical 2D CNNs were employed for liver ROI segmentation: U-Net and Attention U-Net. ROI spectrum signals were normalized and augmented using a sliding window technique. Ultrasound RF signals collected (with a 3-MHz transducer) from 613 participants (Group A) were included for liver ROI segmentation and those from 237 participants (Group B) for liver fibrosis stage classification, with a liver biopsy as the reference standard (Fibrosis stage: F0 = 27, F1 = 49, F2 = 51, F3 = 49, F4 = 61). In the test set of Group A, U-Net and Attention U-Net yielded Dice similarity coefficients of 95.05% and 94.68%, respectively. In the test set of Group B, the 1D CNN performed the best when using ROI phase spectrum signals to evaluate liver fibrosis stages ≥F1 (area under the receive operating characteristic curve, AUC: 0.957; accuracy: 89.19%; sensitivity: 85.17%; specificity: 93.75%), ≥F2 (AUC: 0.808; accuracy: 83.34%; sensitivity: 87.50%; specificity: 78.57%), and ≥F4 (AUC: 0.876; accuracy: 85.71%; sensitivity: 77.78%; specificity: 94.12%), and when using the power spectrum signals to evaluate ≥F3 (AUC: 0.729; accuracy: 77.14%; sensitivity: 77.27%; specificity: 76.92%). The experimental results demonstrated the feasibility of both the 2D and 1D CNNs in liver parenchyma detection and liver fibrosis characterization. The proposed methods have provided a new strategy for liver fibrosis assessment based on ultrasound RF signals, especially for early fibrosis detection. The findings of this study shed light on deep learning analysis of ultrasound RF signals in the frequency domain with automatic ROI segmentation.

Keywords: convolutional neural network; deep learning; liver fibrosis; liver region segmentation; ultrasound radiofrequency signal.

Figure 1

Flow chart of the proposed liver fibrosis stage classification method using deep learning models applied to ultrasound radiofrequency (RF) signals. First, two-dimensional (2D) convolutional neural networks (CNNs) were utilized for automatic liver region of interest (ROI) segmentation from B-mode ultrasound images reconstructed from RF signals. Second, one-dimensional (1D) CNNs were employed to classify liver fibrosis stages based on the frequency spectra (amplitude, phase, and power) of the segmented ROI signals. The white pixels in the liver ROI images represent the detected liver region whose values were 1, and the black pixels represent the detected non-liver region whose values were 0. The ROI spectrum signals were obtained by multiplying the spectrum signals by the resized liver ROI image. The size of the signals or the images is indicated as (a × b), where a is the sampling point number of a frame of signals or the height of an image, and b is the scan line number of the frame of signals or the width of the image.

Figure 2

The network structure of U-Net. Conv: convolution; ReLU: rectified linear unit.

Figure 3

The network structure of Attention U-Net. Conv: convolution; ReLU: rectified linear unit.

Figure 4

The flow chart of data augmentation for ROI spectrum signals. ROI: region of interest.

Figure 5

The network structure of the 1D CNN. Conv: Convolutional layers; 1D: one-dimensional; CNN: convolutional neural network.

Figure 6

Representative liver ROI segmentation results on the test set in Group A by U-Net and Attention U-Net. (a–d) Liver B-mode ultrasound images, (e–h) manual delineation as the reference standard, (i–l) segmentation images by Attention U-Net, (m–p) segmentation images by U-Net. The white pixels in the segmentation images represented the segmented liver region, and the black pixels represented the segmented non-liver region. ROI: region of interest.

Figure 7

Bar chart of the results in Table 1. JSC: Jaccard similarity coefficient; DSC: Dice similarity coefficient; ACC: accuracy; SEN: sensitivity; PRE: precision; SPE: specificity.

Figure 8

Bar chart of the results in Table 3. The compared models were Han et al. [7] and Nguyen et al. [8]. ACC: accuracy; SEN: sensitivity; SPE: specificity; AUC: area under the receiver operating characteristic curve.

Figure 9

Bar chart of the results in Table 4. The compared models were Han et al. [7] and Nguyen et al. [8]. ACC: accuracy; SEN: sensitivity; SPE: specificity; AUC: area under the receiver operating characteristic curve.

Figure 10

Bar chart of the results in Table 5. The compared models were Han et al. [7] and Nguyen et al. [8]. ACC: accuracy; SEN: sensitivity; SPE: specificity; AUC: area under the receiver operating characteristic curve.

Figure 11

ROC curves of the proposed method based on amplitude spectra for classifying liver fibrosis stage ≥F1 (red), ≥F2 (blue), ≥F3 (green), and ≥F4 (purple). ROC: receiver operating characteristic; AUC: area under the ROC curve.

Figure 12

ROC curves of the proposed method based on phase spectra for classifying liver fibrosis stage ≥F1 (red), ≥F2 (blue), ≥F3 (green), and ≥F4 (purple). ROC: receiver operating characteristic; AUC: area under the ROC curve.

Figure 13

ROC curves of the proposed method based on power spectra for classifying liver fibrosis stage ≥F1 (red), ≥F2 (blue), ≥F3 (green), and ≥F4 (purple). ROC: receiver operating characteristic; AUC: area under the ROC curve.