Scatter rejection and scatter correction for single and multiple exposure static diagnostic CT

Static CT designs may offer several advantages, including increased compactness and mechanical simplification. In addition, acquisition speed is no longer constrained by gantry rotation but instead determined by electronic source switching and source power. Static CT further enables multiple simultaneous exposures, for example, in dual-source or even triple-source mode. Due to the static geometry, conventional anti-scatter grids (ASGs) cannot be employed, making scattered radiation a major challenge.

To investigate how 4th generation geometry static CTs can be made suitable for general purpose diagnostic CT imaging with regard to their scattered radiation characteristics.

In this work, we investigate how scatter can be addressed, for example, by using defocused ASGs for scatter rejection and scatter correction approaches. Our evaluation focuses not only on the correction of scatter-induced artifacts but also on the overall image quality, particularly with respect to noise degradation caused by scattered radiation. To do so, we simulate a 4th generation geometry static CT with different levels of collimation in single-source and triple-source exposure mode. It is equipped with a 360 ${}^{\circ }$ detector ring and a 360 ${}^{\circ }$ source ring array, eliminating the need for rotating components. For scatter rejection, we investigate the potential benefits of one-dimensional ASGs in the z-direction by comparing image quality across different ASG heights and phantom sizes. For scatter correction, we employ the deep scatter estimation (DSE), which is based on a U-Net architecture and trained with Monte Carlo-generated scatter estimates. Performance is evaluated with respect to residual scatter artifacts and noise amplification.

DSE effectively corrects scatter artifacts in both single-source and multi-source operation, yielding reconstructions in which scatter artifacts are strongly reduced. However, in multi-source operation we observe a notable increase in image noise at constant dose levels. The use of a one-dimensional ASG in the z-direction can mitigate this noise increase and thereby enable dose savings.

Deep learning-based methods such as DSE can effectively address scatter artifacts. However, the absence of an ASG results in increased image noise. Incorporating a one-dimensional ASG in the z-direction can reduce noise and thereby may enable dose savings.