Low-count PET image reconstruction with Bayesian inference over a Deep Prior

Carrillo, Hernan; Millardet, Maël; Carlier, Thomas; Mateus, Diana

doi:10.1117/12.2580169

Cited by 4 publications

(3 citation statements)

References 0 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Examples include using a prior for the network parameters (and even using posterior distributions to quantify and make use of the uncertainty 70,71 ) and use of stochastic gradient Langevin dynamics (SGLD) for fully Bayesian posterior inference via the DIP. 72,73 A further approach has been Stein’s unbiased risk estimator (SURE) as the loss function for training the DIP (DIP-SURE) instead of using an MSE loss, again to avoid the overfitting problem. 74…”

Section: Ai Methods Without Training Datamentioning

confidence: 99%

AI for PET image reconstruction

Reader

Pan

2023

The British Journal of Radiology

View full text Add to dashboard Cite

Image reconstruction for positron emission tomography (PET) has been developed over many decades, with advances coming from improved modelling of the data statistics and improved modelling of the imaging physics. However, high noise and limited spatial resolution have remained issues in PET imaging, and state-of-the-art PET reconstruction has started to exploit other medical imaging modalities (such as MRI) to assist in noise reduction and enhancement of PET’s spatial resolution. Nonetheless, there is an ongoing drive towards not only improving image quality, but also reducing the injected radiation dose and reducing scanning times. While the arrival of new PET scanners (such as total body PET) is helping, there is always a need to improve reconstructed image quality due to the time and count limited imaging conditions. Artificial intelligence (AI) methods are now at the frontier of research for PET image reconstruction. While AI can learn the imaging physics as well as the noise in the data (when given sufficient examples), one of the most common uses of AI arises from exploiting databases of high-quality reference examples, to provide advanced noise compensation and resolution recovery. There are three main AI reconstruction approaches: i) direct data-driven AI methods which rely on supervised learning from reference data, ii) iterative (unrolled) methods which combine our physics and statistical models with AI learning from data, and iii) methods which exploit AI with our known models, but crucially can offer benefits even in the absence of any example training data whatsoever. This article reviews these methods, considering opportunities and challenges of AI for PET reconstruction.

show abstract

Section: Ai Methods Without Training Datamentioning

confidence: 99%

AI for PET image reconstruction

Reader

Pan

2023

The British Journal of Radiology

View full text Add to dashboard Cite

show abstract

“…Unfortunately, MCMC has a comparable computational cost to Hessian-based approaches and their use is restricted to geophysical tomography [16,17], which has fewer time constraints. A potentially cheaper technique is the stochastic gradient Langevin dynamics algorithm, which has been applied to positron emission tomography reconstruction [18]. Stochastic gradient Langevin dynamics couples stochastic gradient descent to a normally distributed Monte-Carlo sampler.…”

Section: The Reviewmentioning

confidence: 99%

A probabilistic approach to tomography and adjoint state methods, with an application to full waveform inversion in medical ultrasound

et al. 2022

View full text Add to dashboard Cite

Bayesian methods are a popular research direction for inverse problems. There are a variety of techniques available to solve Bayes’ equation, each with their own strengths and limitations. Here, we discuss stochastic variational inference (SVI), which solves Bayes’ equation using gradient-based methods. This is important for applications which are time-limited (e.g. medical tomography) or where solving the forward problem is expensive (e.g. adjoint methods). To evaluate the use of SVI in both these contexts, we apply it to ultrasound tomography of the brain using full-waveform inversion (FWI). FWI is a computationally expensive adjoint method for solving the ultrasound tomography inverse problem, and we demonstrate that SVI can be used to find a no-cost estimate of the pixel-wise variance of the sound-speed distribution using a mean-field Gaussian approximation. In other words, we show experimentally that it is possible to estimate the pixel-wise uncertainty of the sound-speed reconstruction using SVI and a common approximation which is already implicit in other types of iterative reconstruction. Uncertainty estimates have a variety of uses in adjoint methods and tomography. As an illustrative example, we focus on the use of uncertainty for image quality assessment. This application is not limiting; our variance estimator has effectively no computational cost and we expect that it will have applications in fields such as non-destructive testing or aircraft component design where uncertainties may not be routinely estimated.

show abstract

“…But DIP is vulnerable to overfitting and therefore requires manual intervention in early stopping or model under-parametrization. Consequently, a Bayesian neural network (BNN) approach to DIP has been considered to automate the prevention of overfitting [ 22 , 23 ]. Outperforming previous approaches, Posterior Temperature Optimized Bayesian Inverse Models (POTOBIM) have recently shown the most successful application of Bayesian DIP in the context of sparse-view reconstruction [ 24 ].…”

Section: Introductionmentioning

confidence: 99%

Bayesian Reconstruction Algorithms for Low-Dose Computed Tomography Are Not Yet Suitable in Clinical Context

Kniep,

Mieling,

Gerling

et al. 2023

J. Imaging

View full text Add to dashboard Cite

Computed tomography (CT) is a widely used examination technique that usually requires a compromise between image quality and radiation exposure. Reconstruction algorithms aim to reduce radiation exposure while maintaining comparable image quality. Recently, unsupervised deep learning methods have been proposed for this purpose. In this study, a promising sparse-view reconstruction method (posterior temperature optimized Bayesian inverse model; POTOBIM) is tested for its clinical applicability. For this study, 17 whole-body CTs of deceased were performed. In addition to POTOBIM, reconstruction was performed using filtered back projection (FBP). An evaluation was conducted by simulating sinograms and comparing the reconstruction with the original CT slice for each case. A quantitative analysis was performed using peak signal-to-noise ratio (PSNR) and structural similarity index measure (SSIM). The quality was assessed visually using a modified Ludewig’s scale. In the qualitative evaluation, POTOBIM was rated worse than the reference images in most cases. A partially equivalent image quality could only be achieved with 80 projections per rotation. Quantitatively, POTOBIM does not seem to benefit from more than 60 projections. Although deep learning methods seem suitable to produce better image quality, the investigated algorithm (POTOBIM) is not yet suitable for clinical routine.

show abstract

Low-count PET image reconstruction with Bayesian inference over a Deep Prior

Cited by 4 publications

References 0 publications

AI for PET image reconstruction

AI for PET image reconstruction

A probabilistic approach to tomography and adjoint state methods, with an application to full waveform inversion in medical ultrasound

Bayesian Reconstruction Algorithms for Low-Dose Computed Tomography Are Not Yet Suitable in Clinical Context

Contact Info

Product

Resources

About