Breast mass classification based on supervised contrastive learning and multi‐view consistency penalty on mammography

Abstract

Breast cancer accounts for the largest number of patients among all cancers in the world. Intervention treatment for early breast cancer can dramatically extend a woman's 5‐year survival rate. However, the lack of public available breast mammography databases in the field of Computer‐aided Diagnosis and the insufficient feature extraction ability from breast mammography limit the diagnostic performance of breast cancer. In this paper, A novel classification algorithm based on Convolutional Neural Network (CNN) is proposed to improve the diagnostic performance for breast cancer on mammography. A multi‐view network is designed to extract the complementary information between the Craniocaudal (CC) and Mediolateral Oblique (MLO) mammographic views of a breast mass. For the different predictions of the features extracted from the CC view and MLO view of the same breast mass, the proposed algorithm forces the network to extract the consistent features from the two views by the cross‐entropy function with an added consistent penalty term. To exploit the discriminative features from the insufficient mammographic images, the authors learnt an encoder in the classification model to learn the invariable representations from the mammographic breast mass by Supervised Contrastive Learning (SCL) to weaken the side effect of colour jitter and illumination of mammographic breast mass on image quality degradation. The experimental results of all the classification algorithms mentioned in this paper on Digital Database for Screening Mammography (DDSM) illustrate that the proposed algorithm greatly improves the classification performance and diagnostic speed of mammographic breast mass, which is of great significance for breast cancer diagnosis.

Enhancing downstream classification of breast abnormalities in contrast enhanced spectral mammography using a neighborhood representation loss

“…Yan et al proposes an ROI-level contrastive and classification loss for a combined breast mass detection and classification task in full DM images 10 . Similar approaches have also been used by Sun et al and You et al, where a multi-view contrastive loss in addition to the classification loss is suggested 9,11 . However, none of the works has explored the advantages of using SSL-guided classification for CESM ROI classifications.…”

“…Alternatively, contrastive learning-based methods which leverage class label information tackle the problem of learning meaningful representations while mitigating the challenge of data availability 7 . Multiple works have explored various SSL and supervised contrastive learning methods in digital mammography analysis [8][9][10][11] . For example, Cao et al presents an SCL method in multi-view DM images for BIRADS analysis 8 .…”

Improving the malignancy prediction of breast cancer based on the integration of radiomics features from dual-view mammography and clinical parameters

“…Moreover, other studies have shown that precancerous mammograms contain unique imaging information beyond breast density that can be used for risk assessment of breast cancer (Hinton et al 2019;Yala et al 2019;Dembrower et al 2020;Yala et al 2021). Sun et al extracted the complementary information between CC and MLO mammographic views of a breast mass, greatly improving the classi cation performance and diagnostic speed of mammographic breast mass (Sun et al 2022). In addition to their utility in risk assessment, parenchymal and texture features can help evaluate and predict breast cancer (Peter et al 2019).…”