High-resolution MRI synthesis using a data-driven framework with denoising diffusion probabilistic modeling

Chih-Wei Chang,Junbo Peng,Mojtaba Safari,Elahheh Salari,Shaoyan Pan,Justin Roper,Richard L J Qiu,Yuan Gao,Hui-Kuo Shu,Hui Mao,Xiaofeng Yang
DOI: https://doi.org/10.1088/1361-6560/ad209c
IF: 3.5
2024-01-21
Physics in Medicine and Biology
Abstract:High-resolution magnetic resonance imaging (MRI) can enhance lesion diagnosis, prognosis, and delineation. However, gradient power and hardware limitations prohibit recording thin slices or sub-1 mm resolution. Furthermore, long scan time is not clinically acceptable. Conventional high-resolution images generated using statistical or analytical methods include the limitation of capturing complex, high-dimensional image data with intricate patterns and structures. This study aims to harness cutting-edge diffusion probabilistic deep learning techniques to create a framework for generating high-resolution MRI from low-resolution counterparts, improving the denoising diffusion probabilistic model (DDPM) by minimizing its unpredictability and uncertainty. DDPM includes two processes. The forward process employs a Markov chain to systematically introduce Gaussian noise to low-resolution MRI images. In the reverse process, a U-Net model is trained to denoise the forward process images and produce high-resolution images conditioned on the features of their low-resolution counterparts. The proposed framework was demonstrated using T2-weighted MRI images from institutional prostate patients and brain patients collected in the Brain Tumor Segmentation Challenge 2020 (BraTS2020). For the prostate dataset, the bicubic interpolation model (Bicubic), conditional generative-adversarial network (CGAN), and our proposed DDPM framework improved the noise quality measure from low-resolution images by 4.4%, 5.7%, and 12.8%, respectively. Our method enhanced the signal-to-noise ratios by 11.7%, surpassing Bicubic (9.8%) and CGAN (8.1%). In the BraTS2020 dataset, the proposed framework and Bicubic enhanced PSNR from resolution-degraded images by 9.1% and 5.8%. The multi-scale structural similarity indexes were 0.970±0.019, 0.968±0.022, and 0.967±0.023 for the proposed method, CGAN, and Bicubic, respectively. This study explores a deep learning-based diffusion probabilistic framework for improving MR image resolution. Such a framework can be used to improve clinical workflow by obtaining high-resolution images without penalty of the long scan time. Future investigation will likely focus on prospectively testing the efficacy of this framework with different clinical indications.
engineering, biomedical,radiology, nuclear medicine & medical imaging
What problem does this paper attempt to address?
The problem that this paper attempts to solve is: to generate high - resolution magnetic resonance imaging (MRI) images through a data - driven framework and the denoising diffusion probabilistic model (DDPM) without increasing the scanning time or making additional requirements on the hardware. Specifically: 1. **The need for improved resolution**: High - resolution MRI can enhance lesion diagnosis, prognosis, and contour delineation, but is limited by gradient power and hardware limitations, making it difficult to record images with thin - layer or sub - millimeter resolution. In addition, long - time scanning is clinically unacceptable. 2. **Limitations of traditional methods**: Traditional high - resolution image generation methods (such as statistical or analytical methods) cannot capture complex high - dimensional image data and its fine structures and patterns. These methods are usually based on Tikhonov regularization, assuming that there are small local homogeneous regions in high - resolution images, but this assumption does not hold when dealing with complex organs. 3. **Advantages of deep learning**: Compared with traditional methods, data - driven deep - learning methods can capture complex patterns and structures and generate highly realistic images. However, methods based on generative adversarial networks (GAN) may have problems such as training instability or model collapse. To solve these problems, this research proposes a data - driven framework based on the denoising diffusion probabilistic model (DDPM), aiming to generate high - resolution images from low - resolution MRI images and improve the uncertainty problem of DDPM. This framework includes two processes: - **Forward diffusion process**: Systematically introduce Gaussian noise to low - resolution MRI images through a Markov chain. - **Reverse denoising process**: Use the U - Net model for training to remove the noise in the forward process and generate high - resolution images. Through this method, researchers hope to significantly improve the resolution and quality of MRI images without increasing the scanning time, thereby improving the clinical workflow, especially in the application of radiotherapy.