metadata

license: apache-2.0
pipeline_tag: image-feature-extraction
tags:
  - medical
  - cardiac MRI
  - MRI
  - CINE
  - dynamic MRI
  - representation learning
  - unsupervised learning
  - 3D
  - diffusion
  - diffusion autoencoder
  - autoencoder
  - DiffAE
  - 3D DiffAE
  - UK Biobank
  - latent space
library_name: pytorch

UKBBLatent_Cardiac_20208_DiffAE3D_L128_S1701

Biobank-scale imaging provides a unique opportunity to characterise structural and functional cardiac phenotypes and how they relate to disease outcomes. However, deriving specific phenotypes from MRI data requires time-consuming expert annotation, limiting scalability and does not exploit how information dense such image acquisitions are. In this study, we applied a 3D diffusion autoencoder to temporally resolved cardiac MRI data from 71,021 UK Biobank participants to derive latent phenotypes representing the human heart in motion. These phenotypes were reproducible, heritable (h2 = [4 - 18%]), and significantly associated with cardiometabolic traits and outcomes, including atrial fibrillation (P = 8.5 × 10-29) and myocardial infarction (P = 3.7 × 10-12). By using latent space manipulation techniques, we directly interpreted and visualised what specific latent phenotypes were capturing in a given MRI.

Model Details

During this research, the original DiffAE model was adapted and extended for 3D to create the 3D DiffAE model, and was trained on the CINE Cardiac Long-axis 4-chamber view MRIs from UK Biobank dataset using 5 different seeds. This model can be used to infer latent representations from similar cardiac MRIs, or can also be used as pretrained models and then fine-tuned on other datasets or tasks. This model can also be used to generate synthetic cardiac MRIs similar to the training set.

Model Description

Model type: 3D DiffAE
Task: Obtaining latent representation from 3D input volumes
Training dataset: CINE Cardiac Long-axis 4-chamber view MRIs from UK Biobank
Training seed: 1701
Input: 3D MRI (2D over time), intensity normalised (min-max, followed by z-score with 0.5 mean and std)
Output: 128 latent factors. Can also be used for generating synthetic MRIs.

Model Sources

Repository: https://github.com/GlastonburyGroup/ImLatent
Project page: https://glastonburygroup.github.io/CardiacDiffAE_GWAS/
Preprint: https://doi.org/10.1101/2024.11.04.24316700

Citation

If you use this model in your research, or utilise code from this repository or the provided weights, please consider citing the following in your publications:

BibTeX:

@article{Ometto2024.11.04.24316700,
            author       = {Ometto, Sara and Chatterjee, Soumick and Vergani, Andrea Mario and Landini, Arianna and Sharapov, Sodbo and Giacopuzzi, Edoardo and Visconti, Alessia and Bianchi, Emanuele and Santonastaso, Federica and Soda, Emanuel M and Cisternino, Francesco and Ieva, Francesca and Di Angelantonio, Emanuele and Pirastu, Nicola and Glastonbury, Craig A},
            title        = {Unsupervised cardiac MRI phenotyping with 3D diffusion autoencoders reveals novel genetic insights},
            elocation-id = {2024.11.04.24316700},
            year         = {2024},
            doi          = {10.1101/2024.11.04.24316700},
            publisher    = {Cold Spring Harbor Laboratory Press},
            url          = {https://www.medrxiv.org/content/early/2024/11/05/2024.11.04.24316700},
            journal      = {medRxiv}
          }

APA:

Ometto, S., Chatterjee, S., Vergani, A. M., Landini, A., Sharapov, S., Giacopuzzi, E., … Glastonbury, C. A. (2024). Unsupervised cardiac MRI phenotyping with 3D diffusion autoencoders reveals novel genetic insights. medRxiv. doi:10.1101/2024.11.04.24316700