CT Reconstruction using Diffusion Posterior Sampling conditioned on a Nonlinear Measurement Model

Shudong Li,Xiao Jiang,Matthew Tivnan,Grace J. Gang,Yuan Shen,J. Webster Stayman

2024-06-11

Abstract:Diffusion models have been demonstrated as powerful deep learning tools for image generation in CT reconstruction and restoration. Recently, diffusion posterior sampling, where a score-based diffusion prior is combined with a likelihood model, has been used to produce high quality CT images given low-quality measurements. This technique is attractive since it permits a one-time, unsupervised training of a CT prior; which can then be incorporated with an arbitrary data model. However, current methods rely on a linear model of x-ray CT physics to reconstruct or restore images. While it is common to linearize the transmission tomography reconstruction problem, this is an approximation to the true and inherently nonlinear forward model. We propose a new method that solves the inverse problem of nonlinear CT image reconstruction via diffusion posterior sampling. We implement a traditional unconditional diffusion model by training a prior score function estimator, and apply Bayes rule to combine this prior with a measurement likelihood score function derived from the nonlinear physical model to arrive at a posterior score function that can be used to sample the reverse-time diffusion process. This plug-and-play method allows incorporation of a diffusion-based prior with generalized nonlinear CT image reconstruction into multiple CT system designs with different forward models, without the need for any additional training. We develop the algorithm that performs this reconstruction, including an ordered-subsets variant for accelerated processing and demonstrate the technique in both fully sampled low dose data and sparse-view geometries using a single unsupervised training of the prior.

Medical Physics,Computer Vision and Pattern Recognition,Image and Video Processing,Computational Physics

What problem does this paper attempt to address?

The problem that this paper attempts to solve is how to use Diffusion Posterior Sampling (DPS) combined with a nonlinear measurement model to improve image quality in CT image reconstruction. Traditional methods usually rely on a linearized X - ray CT physical model for image reconstruction or restoration, but this linearization is an approximation of the real and essentially nonlinear forward model. This approximation may lead to an inaccurate description of complex imaging systems, resulting in problems in cases such as low - dose data, beam hardening and detector blurring.

To this end, the paper proposes a new method - Diffusion Posterior Sampling based on a nonlinear measurement model (DPS Nonlinear) to solve the inverse problem of nonlinear CT image reconstruction. This method estimates the prior distribution of CT images by training a traditional unconditional diffusion model and applies Bayes' rule to combine this prior with the measurement likelihood score function derived from the nonlinear physical model to obtain the posterior score function for DPS. This method allows the use of diffusion - based priors in combination with multiple forward models in different CT system designs without additional training.

Specifically, the paper addresses the following key issues:
1. **Integration of nonlinear measurement models**: Most existing DPS methods are based on linear measurement models, while the method proposed in this paper can handle nonlinear physical measurement models and more accurately reflect the actual CT imaging process.
2. **No retraining required**: This method can be applied in different CT system measurement models without additional training, improving the flexibility and generality of the method.
3. **Performance improvement at low doses and sparse views**: Verified by experiments, this method performs well under low - dose and sparse - view conditions and can generate high - quality CT images.

The paper evaluates the performance of this method in different scenarios through a series of experiments, including fully - sampled low - dose CT reconstruction, sparse - view CT reconstruction and CT reconstruction under different noise levels. The experimental results show that DPS Nonlinear can effectively improve image quality and reduce errors in these scenarios.