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.

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.

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:

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