Liver Fibrosis Classification based on Multimodal Imaging Feature Fusion

This study introduces a liver fibrosis staging method based on the fusion of multi-modal imaging features. By leveraging ultrasound gray-scale images and ultrasound shear wave elastography (SWE), the method employs an attention-based weighted strategy to effectively fuse different feature maps, integrating information from diverse modalities. Additionally, it integrates multi-modal network branches and designs a multi- modal integrated loss function to update network parameters, thereby enhancing the model's generalization and anti- interference capabilities. The experimental results demonstrate that the proposed fusion network achieves a high AUC of 95.01 and an accuracy of 76.21%. Compared to existing liver fibrosis classification methods for the five-class classification task, which integrate multi-modal features from various liver imaging modalities, our approach shows a significant improvement of 6% in accuracy. With its lightweight model architecture and low computational resource consumption, the proposed method effectively performs liver fibrosis staging, holding significant promise for clinical auxiliary diagnosis.