Non-invasive in vivo histology in health and disease using Magnetic Resonance Imaging (hMRI)
hMRI: extracting histological information from MRI acquisitions
The aim of the ERC-funded histology-MRI (hMRI) project, shown schematically in the image below, is to be able to extract detailed information about the cortex of the human brain, presently available only from invasive ex vivo histology, in vivo using non-invasive MRI. Potential applications of hMRI are numerous; as just two examples, in vivo histology may allow clinicians to diagnose several diseases that can currently only be confirmed post-mortem through ex vivo histology, and neuroscientists to directly probe human cortical networks more directly than currently possible.
hMRI requires a closely intertwined combination of state-of-the-art MRI techniques, biophysical models, and image processing.
1) MRI acquisition techniques.
Development of MRI acquisition is needed to obtain the best data possible. This will be aided by the development of new quantitative MRI acquisitions and also by top of the line hardware, including a Siemens 7T ultra-high field scanner, a Siemens 3T CONNECTOM scanner with 300 mT/m gradients, Skope magnetic field cameras, and a Kineticor prospective motion system.
2) Biophysical models.
Fundamental limits on the resolution of in vivo MRI acquisitions mean that cortical microstructure such as neurons cannot be imaged directly. Instead, we must use biophysical models to relate measured MRI signals to the underlying average microstructure; inversion of these models then allows the extraction of histological information from the MRI data needed for hMRI. A fundamental arm of the project is thus the development and application of appropriate biophysical models.
3) Advanced image processing and machine learning techniques.
The last branch of the project is advanced image processing. This is needed both to improve the acquired data, as well as to allow comparison between hMRI and classical histology for validation of biophysical models. The image processing techniques that we are developing include machine learning-based methods, which have proven very useful for robust analysis of large amounts of data; and super-resolution techniques, which improve the trade-off between spatial resolution, signal-to-noise ratio (SNR) and acquisition time - a necessity for low SNR modalities with long acquisition times such as diffusion-weighted MRI.
Histology-MRI (hMRI) will not be easy to achieve. However, with the rapid developments coming from the rest of the MRI community as well as ourselves, we are confident that it will become possible in the near future. We will all then be able to reap the rewards: Structural information about the human brain currently only available under a microscope will be accessible in vivo and allow revolutions in the study of both neuropathology and neuroscience.