Deep Learning and Structured Prediction for the Segmentation of Mass in Mammograms

Abstract

In this paper, we explore the use of deep convolution and deep belief networks as potential functions in structured prediction models for the segmentation of breast masses from mammograms. In particular, the structured prediction models are estimated with loss minimization parameter learning algorithms, representing: a) conditional random field (CRF), and b) structured support vector machine (SSVM). For the CRF model, we use the inference algorithm based on tree re-weighted belief propagation with truncated fitting training, and for the SSVM model the inference is based on graph cuts with maximum margin training. We show empirically the importance of deep learning methods in producing state-of-the-art results for both structured prediction models. In addition, we show that our methods produce results that can be considered the best results to date on DDSM-BCRP and INbreast databases. Finally, we show that the CRF model is significantly faster than SSVM, both in terms of inference and training time, which suggests an advantage of CRF models when combined with deep learning potential functions.

DCAN: Deep Contour-Aware Networks for Accurate Gland Segmentation

“…Although these methods achieved promising results in cases of adenoma and well differentiated (low grade) adenocarcinoma, they may fail to achieve satisfying performance in malignant subjects, where the glandular structures are seriously deformed. Recently, deep neural networks are driving advances in image recognition related tasks in computer vision [21,9,7,27,29,3] and medical image computing [10,30,31,6,12]. The most relevant study to our work is the U-net that designed a U-shaped deep convolutional network for biomedical image segmentation and won several grand challenges recently [30].…”

A deep learning method for classifying mammographic breast density categories

“…In addition, unlabeled imaging data and unsupervised feature learning (e.g., sparse autoencoder) have been explored for breast density segmentation and risk scoring . Deep convolution and deep belief networks have been integrated in structured prediction models for mammographic breast mass segmentation …”

“…18 Deep learning coupled with big training data has shown promising capability in many artificial intelligence applications 19,20 and, more recently, biomedical imaging analysis. For example, deep learning has been used for thoraco-abdominal lymph node detection and interstitial lung disease classification, 21 realtime 2D/3D registration of digitally reconstructed X-ray images, 22 breast lesion detection and diagnosis, [23][24][25][26][27] radiological imaging segmentation, 28,29 as well as digital breast pathology image analysis 27,30 such as mitosis detection and counting, tissue classification (e.g., cancerous vs. non-cancerous), segmentation (e.g., nuclei or epithelium), 30 and metastatic cancer detection from whole slide images of sentinel lymph nodes. 27 Many such studies have shown that automatic feature extraction using deep learning outperformed traditional hand-crafted imaging descriptors.…”

“…28 Deep convolution and deep belief networks have been integrated in structured prediction models for mammographic breast mass segmentation. 29 In this paper, we propose a deep learning-based approach using CNN to build a computerized reader for BI-RADSbased breast density categorization. Our work focused on distinguishing between the two most difficult to distinguish BI-RADS density categories, i.e., "scattered density" vs. "heterogeneously dense".…”

Adversarial Deep Structural Networks for Mammographic Mass Segmentation

“…The second is the poor contrast of the mammographic images which makes the model predict blurry boundaries and yields inaccurate predictions inside the boundaries. Some works use the structural learning to cope with this difficulty [7,28]. However, the two-stage training used in these works cannot fully explore the power of potential functions.…”