Neural Denoising of Ultra-low Dose Mammography

Abstract

X-ray mammography is commonly used for breast cancer screening. Radiation exposure during mammography restricts the screening frequency and minimal age. Reduction of radiation dose decreases image quality. Image denoising has been recently considered as a way to facilitate dose reduction in mammography without impacting its diagnostic value. We propose a convolutional locally-consistent non-local means (CLC-NLM) algorithm for ultra-low dose mammography denoising. The proposed method achieves powerful denoising while preserving fine details in high resolution mammography. Validation is performed using a dataset of 16 digital mammography cases (4-views each). Since obtaining true low-dose and high-dose mammogram pairs raises regulatory concerns, we applied the X-ray specific and validated method of Veldkamp et al. to simulate 90% dose reduction. The proposed algorithm is shown to compete favorably, both quantitatively and qualitatively, against stateof-the-art neural denoising algorithms. In particular, tiny micro-calcifications are better preserved using the proposed algorithm.

Model-based deep learning to restore low-dose digital breast tomosynthesis images

“…In the field of mammography, it is common to use breast specimen Liu et al (2018), physical phantoms Gao, Fessler and Chan (2021) or virtual clinical trials (VCT) software (SAHU et al, 2019) to generate low-dose and full-dose image pairs for the training process of these deep networks. In Green et al (2019) the authors adapted a noise injection technique from digital chest radiography to simulate ultra-low-dose mammography acquisitions and thus trained a CNN for denoising.…”

“…These works mainly focused on the projection domain, i.e., applying those techniques to restore projection images. Similarly, Green et al (2019) proposed a CNN to restore LD FFDM images. Although they reached very promising results for DBT compared to traditional methods, some limitations include the fact that the networks were trained with breast specimens or physical/simulated phantoms.…”

“…Although they reached very promising results for DBT compared to traditional methods, some limitations include the fact that the networks were trained with breast specimens or physical/simulated phantoms. Different from the other, for FFDM, Green et al (2019) adapted a noise injection technique from digital chest radiography. Although the restoration process in these works depends mainly on the provided data for network learning, incorporating prior knowledge into the network's architecture could potentially enhance its performance.…”