Generating Synthetic Diffusion MRI Maps via Diffusion Denoising Probabilistic Models
Poster No:
1920
Submission Type:
Abstract Submission
Authors:
Tamoghna Chattopadhyay1, Chirag Jagad1, Saket Ozarkar1, Sophia Thomopoulos1, Julio Villalón-Reina1, Paul Thompson1
Institutions:
1University of Southern California, Los Angeles, CA
First Author:
Co-Author(s):
Introduction:
Generative AI has revolutionized synthetic image creation, providing highly realistic outputs for simulations, education, and augmenting datasets in research and clinical domains. Diffusion MRI (dMRI) datasets, especially from large populations (N>10,000), are often limited by high costs and consent for scans that have a relatively long acquisition time. Synthetic imaging offers a feasible alternative by generating realistic neuroimaging data [1,2,3] to supplement existing datasets, facilitating the development of diagnostic tools, simulations, and educational applications. Models such as Generative Adversarial Networks (GANs) and variational autoencoders (VAEs) are commonly used in synthetic neuroimaging. However, GANs often suffer from instability and mode collapse, while VAEs, though reliable, generate images with lower fidelity. Denoising diffusion probabilistic models [4] (DDPMs) have recently emerged as a robust alternative, offering stable, high-quality image synthesis suitable for downstream neuroimaging applications.
Methods:
Diffusion models use a stepwise approach to transform noise into realistic image distributions by learning the inverse process of noising through large-scale training. These models are typically more stable than GANs and achieve greater image quality than VAEs, addressing common limitations such as mode collapse. For this preliminary study, we focused on dMRI, specifically the diffusion tensor imaging–mean diffusivity (DTI-MD) map. This metric is widely used to assess brain microstructure, with applications in studying neurodegenerative diseases, edema, and brain injury. The Cam-CAN dataset, comprising 652 healthy controls (mean age: 54.29±18.59 years; 322 F/330 M), was used for training. To accommodate limited data and computational constraints, we optimized model size, incorporated latent diffusion modeling, and developed conditional diffusion models, building on latent diffusion models (LDMs) developed for 3D T1-weighted brain MRI [1], cross-attention mechanisms enabled image generation conditioned on demographic factors such as sex.
Results:
Both DDPMs and LDMs effectively generated realistic, high-quality synthetic neuroimages. Realism and diversity were quantitatively assessed using established metrics. Maximum Mean Discrepancy (MMD) was employed to evaluate the closeness of synthetic data distributions to real data, where lower scores indicate better similarity. Multi-scale Structural Similarity Index (MS-SSIM) was used to measure visual similarity, with higher values indicating improved fidelity. Image diversity within specific classes (male/female) was also analyzed by assessing distributions of synthetic outputs. We generated 500 synthetic images per model, confirming consistent performance across all evaluations.
Conclusions:
Denoising diffusion models can generate high-fidelity synthetic neuroimaging data, addressing limitations of earlier generative methods such as GANs and VAEs. By computing synthetic DTI-MD maps, these models generate realistic anatomical details and diverse outputs that match the training data distribution. This study highlights their capacity to augment existing datasets, enhance diagnostic model training, and provide valuable resources for neuroimaging applications.
Modeling and Analysis Methods:
Diffusion MRI Modeling and Analysis 2
Novel Imaging Acquisition Methods:
Diffusion MRI 1
Keywords:
Machine Learning
WHITE MATTER IMAGING - DTI, HARDI, DSI, ETC
1|2Indicates the priority used for review
Abstract Information
By submitting your proposal, you grant permission for the Organization for Human Brain Mapping (OHBM) to distribute your work in any format, including video, audio print and electronic text through OHBM OnDemand, social media channels, the OHBM website, or other electronic publications and media.
The Open Science Special Interest Group (OSSIG) is introducing a reproducibility challenge for OHBM 2025. This new initiative aims to enhance the reproducibility of scientific results and foster collaborations between labs. Teams will consist of a “source” party and a “reproducing” party, and will be evaluated on the success of their replication, the openness of the source work, and additional deliverables. Click here for more information. Propose your OHBM abstract(s) as source work for future OHBM meetings by selecting one of the following options:
Please indicate below if your study was a "resting state" or "task-activation” study.
Healthy subjects only or patients (note that patient studies may also involve healthy subjects):
Was this research conducted in the United States?
Are you Internal Review Board (IRB) certified? Please note: Failure to have IRB, if applicable will lead to automatic rejection of abstract.
Were any human subjects research approved by the relevant Institutional Review Board or ethics panel? NOTE: Any human subjects studies without IRB approval will be automatically rejected.
Were any animal research approved by the relevant IACUC or other animal research panel? NOTE: Any animal studies without IACUC approval will be automatically rejected.
Please indicate which methods were used in your research:
For human MRI, what field strength scanner do you use?
Which processing packages did you use for your study?
Provide references using APA citation style.
[2] P.D. Tudosiu, W.H.L. Pinaya, P. Ferreira Da Costa, et al. Realistic morphology-preserving generative modelling of the brain. Nat Mach Intell 6, 811–819 (2024).
[3] L. Puglisi, D. C. Alexander, & D. Ravì, (2024, October). Enhancing spatiotemporal disease progression models via latent diffusion and prior knowledge. In International Conference on Medical Image Computing and Computer-Assisted Intervention (pp. 173-183). Cham: Springer Nature Switzerland.
[4] R. Rombach, et al., “High-Resolution Image Synthesis with Latent Diffusion Models,” in CVPR, 2022, pp. 10674–10685.
[5] S. Thomopoulos, et al., “Diffusion MRI Metrics and their relation to Dementia Severity: Effect of Harmonization Approaches,” (2021) In 17th International Symposium on Medical Information Processing and Analysis (Vol. 12088, pp. 166-179). SPIE.
[6] N.J. Dhinagar, et al., (2024). Counterfactual MRI Generation with Denoising Diffusion Models for Interpretable Alzheimer's Disease Effect Detection. EMBC, 2024.
[7] W. Peng et al., “Metadata-Conditioned Generative Models to Synthesize Anatomically-Plausible 3D Brain MRIs,” pp. 1–26, 2023, [Online]. Available: http://arxiv.org/abs/2310.04630.
[8] Z. Dorjsembe, et al. “Three-Dimensional Medical Image Synthesis with Denoising Diffusion Probabilistic Models,” in MIDL, 2022, pp. 2–4.
[9] G. Kwon, et al. “Generation of 3D Brain MRI Using Auto-Encoding Generative Adversarial Networks,” Lect. Notes Comput. Sci. (including Subser. Lect. Notes Artif. Intell. Lect. Notes Bioinformatics), vol. 11766 LNCS, pp. 118–126, 2019.
[10] J. R. Taylor, et al. (2017). The Cambridge Centre for Ageing and Neuroscience (Cam-CAN) data repository: Structural and functional MRI, MEG, and cognitive data from a cross-sectional adult lifespan sample. Neuroimage, 144, 262-269.