DisAsymNet: Disentanglement of Asymmetrical Abnormality on Bilateral Mammograms using Self-adversarial Learning

TL;DR

DisAsymNet is a novel framework that disentangles asymmetrical abnormalities in bilateral mammograms using self-adversarial learning, improving diagnostic tasks and providing interpretable normal mammogram reconstructions.

Contribution

The paper introduces DisAsymNet, a new method employing asymmetrical abnormality transformer guided self-adversarial learning for disentangling abnormalities in mammograms.

Findings

Outperforms existing methods in classification, segmentation, and localization.

Reconstructed normal mammograms offer better interpretability for clinical diagnosis.

Validated on multiple public and in-house datasets.

Abstract

Asymmetry is a crucial characteristic of bilateral mammograms (Bi-MG) when abnormalities are developing. It is widely utilized by radiologists for diagnosis. The question of 'what the symmetrical Bi-MG would look like when the asymmetrical abnormalities have been removed ?' has not yet received strong attention in the development of algorithms on mammograms. Addressing this question could provide valuable insights into mammographic anatomy and aid in diagnostic interpretation. Hence, we propose a novel framework, DisAsymNet, which utilizes asymmetrical abnormality transformer guided self-adversarial learning for disentangling abnormalities and symmetric Bi-MG. At the same time, our proposed method is partially guided by randomly synthesized abnormalities. We conduct experiments on three public and one in-house dataset, and demonstrate that our method outperforms existing methods in abnormality classification, segmentation, and localization tasks. Additionally, reconstructed normal mammograms can provide insights toward better interpretable visual cues for clinical diagnosis. The code will be accessible to the public.