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

Ultrasound detection of nonalcoholic steatohepatitis using convolutional neural networks with dual-branch global-local feature fusion architecture

Nonalcoholic steatohepatitis (NASH) is a contributing factor to liver cancer, with ultrasound B-mode imaging as the first-line diagnostic tool. This study applied deep learning to ultrasound B-scan images for NASH detection and introduced an ultrasound-specific data augmentation (USDA) technique with a dual-branch global-local feature fusion architecture (DG-LFFA) to improve model performance and adaptability across imaging conditions. A total of 137 participants were included. Ultrasound images underwent data augmentation (rotation and USDA) for training and testing convolutional neural networks-AlexNet, Inception V3, VGG16, VGG19, ResNet50, and DenseNet201. Gradient-weighted class activation mapping (Grad-CAM) analyzed model attention patterns, guiding the selection of the optimal backbone for DG-LFFA implementation. The models achieved testing accuracies of 0.81-0.83 with rotation-based data augmentation. Grad-CAM analysis showed that ResNet50 and DenseNet201 exhibited stronger liver attention. When USDA simulated datasets from different imaging conditions, DG-LFFA (based on ResNet50 and DenseNet201) improved accuracy (0.79 to 0.84 and 0.78 to 0.83), recall (0.72 to 0.81 and 0.70 to 0.78), and F1 score (0.80 to 0.84 for both models). In conclusion, deep architectures (ResNet50 and DenseNet201) enable focused analysis of liver regions for NASH detection. Under USDA-simulated imaging variations, the proposed DG-LFFA framework further improves diagnostic performance.

Convolutional neural network classification of ultrasound parametric images based on echo-envelope statistics for the quantitative diagnosis of liver steatosis

PURPOSE

Early detection and quantitative evaluation of liver steatosis are crucial. Therefore, this study investigated a method for classifying ultrasound images to fatty liver grades based on echo-envelope statistics (ES) and convolutional neural network (CNN) analyses.

METHODS

Three fatty liver grades, i.e., normal, mild, and moderate-to-severe, were defined using the thresholds of the magnetic resonance imaging-derived proton density fat fraction (MRI-PDFF). There were 10 cases of each grade, totaling 30 cases. To visualize the texture information affected by the deposition of fat droplets within the liver, the maps of first- and fourth-order moments and the heat maps formed from both moments were employed as parametric images derived from the ES. Several dozen to hundreds of regions of interest (ROIs) were extracted from the liver region in each parametric image. A total of 7680 ROIs were utilized for the transfer learning of a pretrained VGG-16 and classified using the transfer-learned VGG-16.

RESULTS

The classification accuracies of the ROIs in all types of the parametric images were approximately 46%. The fatty liver grade for each case was determined by hard voting on the classified ROIs within the case. In the case of the fourth-order moment maps, the classification accuracy of the cases through hard voting mostly increased to approximately 63%.

CONCLUSIONS

The formation of parametric images derived from the ES and the CNN classification of the parametric images were proposed for the quantitative diagnosis of liver steatosis. In more than 60% of the cases, the fatty liver grade could be estimated solely using ultrasound images.