Breast Ultrasound Image Classification Using Transfer Learning-Based Feature Fusion Mode
Faming Li, Ying Te Wu, Bo Xu, Cuier Tan · 2024
In the field of ultrasound image recognition, the recognition rate is difficult to improve solely using a single model due to the insufficient coverage of feature extraction. This paper presents a transfer learning-based feature fusion model called RVB-Net--composed of the feature extractor group consisting of ResNet101 and VGGNet19, as well as the CNN_B module. The model combines two classic models with significant differences in depth as the feature extractors. The resulting 7x7 feature tensors from their respective feature layers are saved and passed through our designed feature reprocessing module--CNN_B, performing multi-path parallel feature reprocessing to obtain the final classification results. Experimental results demonstrate that the RVB-Net exhibits superior feature extraction capability compared to other fusion models and single models. Additionally, the model demonstrates stronger feature reprocessing capability, achieving the best performance in classification experiments with an accuracy of $\mathbf{9 5. 3 6 \%}$, precision of $\mathbf{9 5. 5 1 \%}$, recall of $\mathbf{9 7. 7 0 \%}$, and F1score of 0.966.