Reconstructed Multi-Scale Feature Fusion Network for Breast Tumor Classification

Early and effective diagnosis is crucial for the treatment of breast cancer. However, existing methods are limited by image quality and often fail to capture the morphological features of complex tumors, resulting in suboptimal diagnostic accuracy. We propose a novel method that integrates image reconstruction and multi-scale feature fusion to enhance the classification accuracy of breast tumors. Specifically, the Super-Resolution Convolutional Neural Network is employed for image reconstruction to improve image quality. Subsequently, multiscale feature extraction and fusion are carried out through the Spatial Pyramid Pooling with Context module to enhance feature representation. Comparative and ablation experiments are conducted on DOBI and BreakHis datasets, the results demonstrate that our method achieves more robust and accurate performance in tumor classification.