Deep learning grading of Schistosoma japonicum-associated liver fibrosis from ultrasound images: a weighted soft-voting ensemble of convolutional and transformer networks

A weighted soft-voting ensemble of ResNet-18, EfficientNet-B0, DenseNet-121, and Swin-T grades Schistosoma japonicum-associated liver fibrosis from single B-mode ultrasound images with macro-F1 0.8527 on 440 test images.

• Why it matters: Ultrasound fibrosis grading for schistosomiasis remains subjective and limited by scarce experienced sonographers in endemic counties, creating a need for more consistent, automated support.
• What they did: Using 2,921 adult ultrasound images from a Jiangxi Province hospital (2019–2023) with fibrosis labels from clinical records and radiologist re-evaluation, the ensemble combined four networks via validation-weighted soft voting.
• The result: For significant fibrosis (grade 2 or 3; prevalence 0.530), it reached sensitivity 0.9700, specificity 0.9179, and AUROC 0.9817, with misclassifications only to adjacent grades.

AI summary of the abstract · may contain errors · not medical advice · read the paper.