Keywords: self-supervised learning, dense breast, pathology aware, CAD
Abstract: Screening mammography is recognized as an effective method to diagnose breast cancer (BC). However, for extremely dense breasts, there is a higher chance to induce misdiagnosing. To suppress misdiagnosis from radiologists in mammography reading, computer-aided diagnosis (CAD) based on imaging has been widely researched and applied. These CAD tools increasingly have deeper layers design aiming for better performance, but this may decrease robustness particularly in dense breast. Therefore, to benefit BC identification in the context of supervision from rare annotated datasets, we propose a self-supervised learning framework to normalize mammograms into pathology aware (PA) style, which is in line with the pathological local enhancement characteristic, and prove the value of PA mammogram for the downstream tasks. Experimental results on INBreast and CBIS-DDSM datasets suggest that our method can achieve better performance in both normal and dense breasts for classification and segmentation tasks.
Registration: I acknowledge that acceptance of this work at MIDL requires at least one of the authors to register and present the work during the conference.
Authorship: I confirm that I am the author of this work and that it has not been submitted to another publication before.
Paper Type: novel methodological ideas without extensive validation
Primary Subject Area: Detection and Diagnosis
Secondary Subject Area: Application: Radiology
Confidentiality And Author Instructions: I read the call for papers and author instructions. I acknowledge that exceeding the page limit and/or altering the latex template can result in desk rejection.