Domain Generalization for Mammography Detection via Multi-style and Multi-view Contrastive Learning

Abstract

Mammographic image analysis is a fundamental problem in the computer-aided diagnosis scheme, which has recently made remarkable progress with the advance of deep learning. However, the construction of a deep learning model requires training data that are large and sufficiently diverse in terms of image style and quality. In particular, the diversity of image style may be majorly attributed to the vendor factor. However, mammogram collection from vendors as many as possible is very expensive and sometimes impractical for laboratory-scale studies. Accordingly, to further augment the generalization capability of deep learning models to various vendors with limited resources, a new contrastive learning scheme is developed. Specifically, the backbone network is firstly trained with a multi-style and multi-view unsupervised selflearning scheme for the embedding of invariant features to various vendor styles. Afterward, the backbone network is then recalibrated to the downstream tasks of mass detection, multi-view mass matching, BI-RADS classification and breast density classification with specific supervised learning. The proposed method is evaluated with mammograms from four vendors and two unseen public datasets. The experimental results suggest that our approach can effectively improve analysis performance on both seen and unseen domains, and outperforms many state-of-the-art (SOTA) generalization methods.

Multi-View Computed Tomography Network for Osteoporosis Classification

“…[17], [18] attempted to integrate contrastive learning and volumetric medical domain-specific knowledge for volumetric medical segmentation. Li et al [19] proposed a self-learning scheme that leverages multi-style and multi-view to reduce the vendor-style domain gap for mammography detection. Yang et al [20] designed a hybrid representation learning which use domain-specific knowledge in histopathological images.…”

Image Augmentation Techniques for Mammogram Analysis

“…The resulting images helped the authors work out the training of a deep CNN for breast cancer classification. A new domain generalization method is proposed in [ 36 ] to aid mammography lesion detection techniques. Domain-invariant features were embedded in a range of datasets using a multi-style and multi-view contrastive learning technique.…”

Researches advanced in application of medical image analysis based on deep learning

“…Therefore, researchers resort to computers to help deal with this difficult task. Due to the rise of deep learning methods, the medical image detection topic is popular even until now and has been applied to detect many targets, for example, lung nodules, organ and lesions in mammograms [5][6][7]. After reviewing the recent advance in medical image detection area, the following two directions are most frequently explored, 3D contextual information incorporation and high-quality data availability, the other directions are put into the third section.…”