Development of scatter correction method using convolutional neural network in digital breast tomosynthesis

Abstract

With the development of flat-panel detector, in X-ray imaging, X-ray scatter causes image quality degradation and incorrect estimation of primary radiation. To reduce or correct scatter artifacts, many kinds of methods have been continuously developed, and among them, the kernel-based scatter correction method has been widely adopted in terms of fast calculation and easy implementation. However, the kernel-based method using pencil beam is less accurate in estimating scatter near the breast boundary in digital breast tomosynthesis, where rotation of the source is added. In this study, we proposed a supervised learning-based scatter correction method using a convolutional neural network in numerical breast phantom with Monte Carlo simulation. Through the proposed method, we confirmed better performance in estimation of scatter than conventional kernel-based method and robustness according to breast density. We expect that the proposed method can be extended to clinical data in the future.