George Webber

Better algorithms for medical image reconstruction can improve image quality and enable reductions in acquisition time and radiation dose. A prior understanding of the distribution of plausible images is key to realising these benefits. Recently, research into deep-learning image reconstruction has started to look into using unsupervised diffusion models, trained only on high-quality medical images (i.e. without needing paired scanner measurement data), for modelling this prior understanding. Image reconstruction algorithms incorporating unsupervised diffusion models have already attained state-of-the-art accuracy for reconstruction tasks ranging from highly accelerated MRI to ultra-sparse-view CT and low-dose PET. Key advantages of diffusion model approach over previous deep learning approaches for reconstruction include state-of-the-art image distribution modelling, improved robustness to domain shift, and principled quantification of reconstruction uncertainty. If hallucination concerns can be alleviated, their key advantages and impressive performance could mean these algorithms are better suited to clinical use than previous deep-learning approaches. In this review, we provide an accessible introduction to image reconstruction and diffusion models, outline guidance for using diffusion-model-based reconstruction methodology, summarise modality-specific challenges, and identify key research themes. We conclude with a discussion of the opportunities and challenges of using diffusion models for medical image reconstruction.

Medical image reconstruction with pretrained score-based generative models (SGMs) has advantages over other existing state-of-the-art deep-learned reconstruction methods, including improved resilience to different scanner setups and advanced image distribution modeling. SGM-based reconstruction has recently been applied to simulated positron emission tomography (PET) datasets, showing improved contrast recovery for out-of-distribution lesions relative to the state-of-the-art. However, existing methods for SGM-based reconstruction from PET data suffer from slow reconstruction, burdensome hyperparameter tuning and slice inconsistency effects (in 3D). In this work, we propose a practical methodology for fully 3D reconstruction that accelerates reconstruction and reduces the number of critical hyperparameters by matching the likelihood of an SGM’s reverse diffusion process to a current iterate of the maximum-likelihood expectation maximization algorithm. Using the example of low-count reconstruction from simulated [¹⁸F]DPA-714 datasets, we show our methodology can match or improve on the NRMSE and SSIM of existing state-of-the-art SGM-based PET reconstruction while reducing reconstruction time and the need for hyperparameter tuning. We evaluate our methodology against state-of-the-art supervised and conventional reconstruction algorithms. Finally, we demonstrate a first-ever implementation of SGM-based reconstruction for real 3D PET data, specifically [¹⁸F]DPA-714 data, where we integrate perpendicular pre-trained SGMs to eliminate slice inconsistency issues.

Diffusion models (DMs) have recently been introduced as a regularizing prior for PET image reconstruction, integrating DMs trained on high-quality PET images with unsupervised schemes that condition on measured data. While these approaches have potential generalization advantages due to their independence from the scanner geometry and the injected activity level, they forgo the opportunity to explicitly model the interaction between the DM prior and noisy measurement data, potentially limiting reconstruction accuracy. To address this, we propose a supervised DM-based algorithm for PET reconstruction. Our method enforces the non-negativity of PET’s Poisson likelihood model and accommodates the wide intensity range of PET images. Through experiments on realistic brain PET phantoms, we demonstrate that our approach outperforms or matches state-of-the-art deep learning-based methods quantitatively across a range of dose levels. We further conduct ablation studies to demonstrate the benefits of the proposed components in our model, as well as its dependence on training data, parameter count, and number of diffusion steps. Additionally, we show that our approach enables more accurate posterior sampling than unsupervised DM-based methods, suggesting improved uncertainty estimation. Finally, we extend our methodology to a practical approach for fully 3D PET and present example results from real [¹⁸F]FDG brain PET data.