获取途径

作者

Changchao Wang Wenzong Jiang Chao Li Zhiyong Yu Weifeng Liu Xiongbin Wang & …Baodi Liu

作者单位

²Shandong First Medical University Affiliated Tumor Hospital, Jinan, 250000, China ³The Second Affiliated Hospital of Kunming Medical University, Kunming, 650000, China ¹College of Control Science and Engineering, China University of Petroleum (East China), Qingdao, 266580, China

刊名

Multimedia Systems

年份

2026

卷号

Vol.32 No.3

ISSN

0942-4962

关键词

Breast cancer Mammography Hard samples Supervised contrastive learning

摘要

Breast cancer is the most common cancer among women worldwide, and its incidence continues to rise. Early detection is crucial for improving prognosis. Mammography remains the gold standard for early breast cancer screening due to its high sensitivity, non-invasive nature, affordability, and ability to detect small lesions. In recent years, some studies have theoretically proven that hard positive/negative samples can enhance the capability of models using supervised contrastive loss. However, t...更多

Breast cancer is the most common cancer among women worldwide, and its incidence continues to rise. Early detection is crucial for improving prognosis. Mammography remains the gold standard for early breast cancer screening due to its high sensitivity, non-invasive nature, affordability, and ability to detect small lesions. In recent years, some studies have theoretically proven that hard positive/negative samples can enhance the capability of models using supervised contrastive loss. However, the construction of hard samples remains a pressing challenge, particularly in mammogram classification. Inspired by radiologists who can quickly enhance their diagnostic skills through exposure to “unusual” cases in clinical practice, this paper proposes a novel method called low-frequency attenuation supervised contrastive learning for mammogram classification . The SupCon-LFA reduces the contextual information of lesion features in mammography images through low frequency attenuation , generating challenging hard positive and hard negative samples. These hard samples can provide crucial gradient information during the training phase, thereby enhancing the model’s ability to recognize lesion types. Experimental results show that SupCon-LFA significantly outperforms other mainstream methods in mammography image classification tasks, demonstrating its potential in enhancing model classification performance. SupCon-LFA provides an effective tool for the accurate screening of early-stage breast cancer.收起

文献类型

期刊