[Paper Review] High-Fidelity Modeling of Detector Lag and Gantry Motion in CT Reconstruction

Steven Tilley, Alejandro Sisniega+2 more|arXiv (Cornell University)|May 29, 2018

Medical Imaging Techniques and Applications14 references4 citations

TL;DR

This paper proposes a high-fidelity forward model integrated into model-based iterative reconstruction (MBIR) to jointly correct for detector lag and gantry motion blur in CT imaging. By modeling both effects within a penalized, weighted nonlinear least-squares framework, the method reduces azimuthal blurring and lag artifacts, significantly improving image quality—especially in low-contrast and peripheral regions—without pre-processing corrections.

ABSTRACT

Detector lag and gantry motion during x-ray exposure and integration both result in azimuthal blurring in CT reconstructions. These effects can degrade image quality both for high-resolution features as well as low-contrast details. In this work we consider a forward model for model-based iterative reconstruction (MBIR) that is sufficiently general to accommodate both of these physical effects. We integrate this forward model in a penalized, weighted, nonlinear least-square style objective function for joint reconstruction and correction of these blur effects. We show that modeling detector lag can reduce/remove the characteristic lag artifacts in head imaging in both a simulation study and physical experiments. Similarly, we show that azimuthal blur ordinarily introduced by gantry motion can be mitigated with proper reconstruction models. In particular, we find the largest image quality improvement at the periphery of the field-of-view where gantry motion artifacts are most pronounced. These experiments illustrate the generality of the underlying forward model, suggesting the potential application in modeling a number of physical effects that are traditionally ignored or mitigated through pre-corrections to measurement data.

Motivation & Objective

• Address the degradation of image quality in high-resolution CT applications due to detector lag and gantry motion blur.
• Overcome limitations of traditional pre-correction methods by integrating physical blur models directly into the reconstruction process.
• Demonstrate that joint modeling of detector lag and gantry motion within a unified MBIR framework improves spatial resolution and reduces artifacts.
• Evaluate the performance of different regularization penalties (Huber vs. quadratic) in deblurring gantry motion effects while preserving edges.
• Establish a generalizable framework applicable to a wide range of physical effects in CT systems beyond just detector lag and motion blur.

Proposed method

• Adopt a general forward model from prior work (Tilley et al., 2018) expressed as $ \bar{\bm{y}} = \bm{B} \exp(-\bm{A} \bm{\mu}) $, where $ \bm{y} \sim \mathcal{N}(\bar{\bm{y}}, \bm{K}) $, allowing flexible incorporation of physical effects.
• Model detector lag using a sum-of-exponentials impulse response: $ h[k] = b_0\delta[k] + \sum_{i=1}^{3} b_i \exp(-k a_i) $ for $ 0 \leq k < K $, representing temporal charge trapping and release.
• Model gantry motion blur as an azimuthal convolution over the projection angle, integrating x-ray intensity over the arc traversed during detector integration time.
• Formulate a penalized, weighted nonlinear least-squares objective: $ \| \bm{y} - \bm{B} \exp(-\bm{A} \bm{\mu}) \|_{\bm{K}^{-1}}^2 + \beta R(\bm{\mu}) $, minimizing this to obtain the reconstructed attenuation map $ \bm{\mu} $.
• Use ordered subsets and Nesterov momentum acceleration to improve convergence speed in iterative reconstruction.
• Compare reconstruction performance using Huber and quadratic penalties, with the Huber penalty shown to better preserve sharp edges during deblurring.

Experimental results

Research questions

• RQ1 Can a unified forward model in MBIR effectively correct for both detector lag and gantry motion blur in CT without pre-processing corrections?
• RQ2 How does modeling detector lag within the reconstruction framework compare to traditional pre-correction methods in reducing characteristic lag artifacts?
• RQ3 To what extent does modeling gantry motion blur improve spatial resolution, particularly in peripheral regions of the field-of-view?
• RQ4 Which regularization penalty (Huber vs. quadratic) enables better deblurring of gantry motion effects while maintaining edge sharpness?
• RQ5 Can the proposed framework be extended to model additional physical effects such as focal spot blur or scintillator blur?

Key findings

• Modeling detector lag within the MBIR forward model successfully eliminated the characteristic bright trail artifact arcing from the skull into the brain in both simulated and physical head phantom studies.
• In physical experiments, the lag correction reduced the artifact significantly, though residual trails suggest potential for improvement via more accurate kernel estimation or increased iterations.
• For gantry motion blur, the proposed model reduced bias in regions far from the isocenter—particularly at 60 mm from center—demonstrating improved spatial resolution where motion blur is most pronounced.
• Reconstructions with the blur model accurately preserved circular features at 100 mm from center, while the identity model (no blur modeling) showed increasing blurring along the direction of rotation.
• The Huber penalty outperformed the quadratic penalty in gantry motion correction by preserving sharp edges and enabling effective deblurring without excessive noise amplification.
• Noise levels were well-matched between models (7.91×10⁻⁵ mm⁻¹ for blur model vs. 7.96×10⁻⁵ mm⁻¹ for identity in detector lag study), confirming that improvements were due to modeling fidelity, not noise level differences.