MR Image Denoising and Super-Resolution Using Regularized Reverse Diffusion

Abstract

Patient scans from MRI often suffer from noise, which hampers the diagnostic capability of such images. As a method to mitigate such artifact, denoising is largely studied both within the medical imaging community and beyond the community as a general subject. However, recent deep neural network-based approaches mostly rely on the minimum mean squared error (MMSE) estimates, which tend to produce a blurred output. Moreover, such models suffer when deployed in real-world sitautions: out-of-distribution data, and complex noise distributions that deviate from the usual parametric noise models. In this work, we propose a new denoising method based on score-based reverse diffusion sampling, which overcomes all the aforementioned drawbacks. Our network, trained only with coronal knee scans, excels even on out-of-distribution in vivo liver MRI data, contaminated with complex mixture of noise. Even more, we propose a method to enhance the resolution of the denoised image with the same network. With extensive experiments, we show that our method establishes state-of-theart performance, while having desirable properties which prior MMSE denoisers did not have: flexibly choosing the extent of denoising, and quantifying uncertainty.

Generative Modelling With Inverse Heat Dissipation

“…The proposed method could effectively be used for superresolution tasks, and the way that the super-resolution is resolved will be affected by biases in the training data, which is something that users need to be aware of. On the other hand, generative models have been applied to inverse problems in medical imaging [32,74,10,12], for instance. Another positive potential is to augment art and human creativity by including manual controls to the synthesis process, e.g., text-conditional generation [59].…”

“…Menick and Kalchbrenner [53] suggest a similar method based on resolution and bit-depth upscaling, although in a non-parallel way. Previously, generative models, including diffusion models, have been utilized for image deblurring [45,46,1,80], super-resolution [48,65,14,57,7,64,12], and other types of inverse problems [38,11,32,74,10,39]. While our model does effectively perform deblurring/superresolution, the main difference to these works is that instead of using a pre-existing generative model to solve the inverse problem, we do the exact opposite and create a new generative model through iteratively solving an inverse problem using a loss function that directly reverses the heat equation.…”

Diffusion Models for Medical Image Analysis: A Comprehensive Survey

“…Conventional deep-based networks utilize Minimum Mean Square Error (MMSE) estimates, which lead to unsatisfactory and blurred images due to the distribution change in train/test data or the preliminary assumption of Gaussian noise is at odds with the data's actual distribution. Chung et al [58] proposed a multi-successive paradigm for MRI image denoising and super-resolution, namely R2D2+, with the SDE [62] algorithm to tackle the mentioned deficiencies. Diffusion generative models are robust to any distribution change over the data and produce more realistic data [23].…”

“…Despite the advantages of diffusion models, they are very time-consuming. To this end, Chung et al [58] do not start the reverse diffusion process from the pure noise but start from the initial noisy image. R2D2+ [58] solves a reverse time SDE procedure with a non-parametric estimation method based on eigenvalue analysis of covariance matrix rather than the conventional numerical methods [62].…”

Unsupervised Medical Image Translation with Adversarial Diffusion Models

“…Due to the gradual stochastic sampling process and explicit likelihood characterization, diffusion models can offer enhanced sample quality and diversity. Given this potential, diffusion-based methods have recently been adopted for medical image reconstruction [45]- [47] and denoising [48] tasks. Yet, these methods are characteristically based on unconditional diffusion processes devised to enhance the quality of single-modality images.…”