Empirical Motion-Artifact Reduction for Non-Rigid Motion in Dedicated Breast CT

The goal of this work is to develop a data-driven empirical motion-artifact reduction algorithm for non-rigid motion in dedicated breast CT.

Breast CT is a novel imaging modality that offers fully 3D images at good spatial resolution without breast compression and tissue overlap. However, the slow rotation speed of the gantry in such systems increases the likelihood of motion artifacts. Because of the breast anatomy, motion-artifact reduction techniques need to be able to handle artifacts induced by non-rigid motion, which cannot be modeled due to variable motion patterns and the breasts' inner structure, shape, and size. In this work, we present an iterative data-driven empirical algorithm to reduce motion artifacts in breast CT. The highlight of our method is the ability to perform transformations in the image domain using b-spline fields that are defined for each angle and can be efficiently updated with gradient descent and automatic differentiation.

We test the method using a simulation study, on physical phantoms, and clinical cases, and show that it can significantly reduce the appearance of motion artifacts.

This work introduces a fully data-driven empirical motion-artifact reduction capable of identifying and minimizing motion artifacts without an underlying model of motion.