A parasitic metric learning net for breast mass classification based on mammography. (March 2018)

• Record Type: Journal Article
• Authors: Jiao, Zhicheng; Gao, Xinbo; Wang, Ying; Li, Jie
• Is Part Of: Pattern recognition. Volume 75(2018:Mar.)
• Page Start: 292
• Page End: 301
• Publication Date: 2018-03
• Subjects: Deep learning -- Metric learning -- Breast mass classification -- Mammography -- Convolutional neural network; Pattern perception -- Periodicals; Perception des structures -- Périodiques; Patroonherkenning; 006.4
• DOI: 10.1016/j.patcog.2017.07.008
• Languages: English
• ISSNs: 0031-3203

Abstract

Highlights: A CNN model is proposed and trained by us to obtain deep representation of breast mass images. A combined model containing metric learning net and CNN is introduced to improve performance on breast mass classification. A novel training strategy is introduced in this task to improve performance of traditional stochastic gradient descent. Abstract: Accurate classification of different tumors in mammography plays a critical role in the early diagnosis of breast cancer. However, owing to variations in appearance, it is a challenging task to distinguish malignant instances from benign ones. For this purpose, we train a deep convolutional neural networks (CNNs) to obtain more discriminative description of breast tissues. Benefiting from the discriminative representation, metric learning layers are proposed to further improve performance of the deep structure. The best-performing model restricts the depth of backpropagation of joint training in only the metric learning layers. Relation between metric learning layers and tradition CNNs structures seems like parasitism relationship between species, where one species, the parasite, benefits at the expense of the other. Therefore, the proposed method is named as parasitic metric learning net. To confirm veracity of our method, classification experiments on breast mass images of two widely used databases are performed. Comparing performance of the proposed method with traditional ones, competitive results are achieved. Meanwhile, the parameter updating strategy for our parasitic metric net may inspire a way of improving performance of a pre-trained CNNs model on particular medical image processing or other computer vision tasks.