DCANet: Dual contextual affinity network for mass segmentation in whole mammograms.

PURPOSE: Breast mass segmentation in mammograms remains a crucial yet challenging topic in computer-aided diagnosis systems. Existing algorithms mainly used mass-centered patches to achieve mass segmentation, which is time-consuming and unstable in clinical diagnosis. Therefore, we aim to directly perform fully automated mass segmentation in whole mammograms with deep learning solutions. METHODS: In this work, we propose a novel dual contextual affinity network (a.k.a., DCANet) for mass segmentation in whole mammograms. Based on the encoder-decoder structure, two lightweight yet effective contextual affinity modules including the global-guided affinity module (GAM) and the local-guided affinity module (LAM) are proposed. The former aggregates the features integrated by all positions and captures long-range contextual dependencies, aiming to enhance the feature representations of homogeneous regions. The latter emphasizes semantic information around each position and exploits contextual affinity based on the local field-of-view, aiming to improve the indistinction among heterogeneous regions. RESULTS: The proposed DCANet is greatly demonstrated on two public mammographic databases including the DDSM and the INbreast, achieving the Dice similarity coefficient (DSC) of 85.95% and 84.65%, respectively. Both segmentation performance and computational efficiency outperform the current state-of-the-art methods. CONCLUSION: According to extensive qualitative and quantitative analyses, we believe that the proposed fully automated approach has sufficient robustness to provide fast and accurate diagnoses for possible clinical breast mass segmentation.

FS-UNet: Mass segmentation in mammograms using an encoder-decoder architecture with feature strengthening.

Breast mass segmentation in mammograms is still a challenging and clinically valuable task. In this paper, we propose an effective and lightweight segmentation model based on convolutional neural networks to automatically segment breast masses in whole mammograms. Specifically, we first developed feature strengthening modules to enhance relevant information about masses and other tissues and improve the representation power of low-resolution feature layers with high-resolution feature maps. Second, we applied a parallel dilated convolution module to capture the features of different scales of masses and fully extract information about the edges and internal texture of the masses. Third, a mutual information loss function was employed to optimise the accuracy of the prediction results by maximising the mutual information between the prediction results and the ground truth. Finally, the proposed model was evaluated on both available INbreast and CBIS-DDSM datasets, and the experimental results indicated that our method achieved excellent segmentation performance in terms of dice coefficient, intersection over union, and sensitivity metrics.