Deep Learning guided Imaging to correlate imaging from a whole organ to cellular level

Brief description of the context of the research including potential impact

Hierarchical Phase-Contrast Tomography (HiP-CT) is an X-ray technique developed using high-energy X-rays, that allow us to create images of whole human organs at high resolution (ca. 1 um) and in three dimensions. This allows us to better understand the complex structure and function of the human body, as well as to better understand changes caused by disease.

Our new imaging technique is similar to the X-ray CT used widely in conventional medical imaging but uses a synchrotron X-ray source based at the ESRF (European Synchrotron Radiation facility in Grenoble). This X-ray source offers the brightest and most coherent beam in the world, and, coupled to the HIP-CT technique we're developing, allows us to image entire human organs (including lung, heart, brain) with 25um resolution, and zoom in on cellular structures at ~1.2um resolution without cutting the tissue. We have imaged human organs in health and disease including Covid-19 victims.

Aims and Objectives The specific objectives are to:

Develop and apply deep learning techniques to segment HiP-CT data (airways, blood vessels, cells, etc.) to enable biological insights to be drawn and for further biophysical simulations.

Explore more advanced machine learning techniques such as generative adversarial networks, in order to correlate HiP-CT data with images from other modalities (such as histology, lightsheet, MRI and CT). This type of analysis will enable substantially better interpretation of HiP-CT so that it can provide quantitative biological and medical insights.

Novelty of Research Methodology Alignment to EPSRC's strategies and research areas Any companies or collaborators involved

The project is an international interdisciplinary collaboration between scientists and mathematicians at UCL, ESRF and DLS, and clinicians at Hannover-biobank, Mainz and Heidelberg, UCL and Imperial College London, together with many other collaborators.