We use techniques from ML, compressed sensing and Bayes’ theorem to improve the quality, speed or cost of medical scans.

We can tackle many different data modalities (brain MRIs, PET, CT, …), chest X-rays, any many more. Such techniques can correct scanning artefacts, motion of patients in the scanner, re-construct from 2D to 3D, and recontruct undersampled MRI which only measure a few points in k-space.

We also use Bayesian techniques such as MCMC or Variational Inference to learn a distribution over the space of potential solutions, this is highly important for such ill-posed problems.

Further reading:

• Marinescu et al, Bayesian Image Reconstruction using Deep Generative Models, 2021
• Mildenhall et al, NeRF: Representing Scenes as Neural Radiance Fields for View Synthesis, 2020
• Zhu et al, Image reconstruction by domain transform manifold learning, 2017
• Sitzmann et al, Implicit Neural Representations with Periodic Activation Functions, 2020