Brain imaging methods : the state of the art

Abstract

In the first lecture, we reviewed the various brain imaging methods and their differences. Functional MRI (fMRI), now the most widely used, involves measuring small variations in cerebral oxygenation which, thanks to a neurovascular coupling mechanism, reflect the activity of neuronal circuits. FMRI offers a number of advantages : rapid, repeated, whole-brain measurements, without the need for injections. Although the measurement is indirect, it reflects neuronal discharges fairly faithfully and with a high degree of linearity (except in certain pathological cases). However, spatial resolution remains modest (of the order of a millimeter or a little less), and only allows us to visualize the average of populations of the order of several tens of thousands of neurons. Spatial resolution is also reduced, although analysis techniques can detect variations of the order of 100-200 milliseconds in the onset and duration of activation.

MRI is constantly progressing, notably with the use of parallel antennas with 32 or 64 channels, multiband sequences, parallel transmission and partial sampling techniques. Increasing the magnetic field improves the signal-to-noise ratio, which varies non-linearly. We have gone from 1.5 Tesla machines in the 1990s to MRIs at 3 T, then 7 T, 9.4 T, and finally, at NeuroSpin, to a unique prototype magnet 90 cm in diameter at 11.72 T, whose field rise was achieved in July 2019. These improvements not only concern functional MRI, but also the analysis of brain anatomy, at both mesoscopic (~1 millimeter) and microscopic (a few microns) scales. Different variants of diffusion MRI make it possible, for example, to assess the average diameter of axons, the average size of cells, and even their shape, without, however, seeing them individually.

Complementary to MRI, electroencephalography (EEG) and magnetoencephalography (MEG) directly measure electrical potentials and tiny magnetic fields generated by currents flowing between the dendrites and cell bodies of neurons. This is why these methods have excellent temporal (sub-millisecond) and frequency resolution (between 0.01 and 1000 Hz or more). Their excellent sensitivity, coupled with the use of multivariate decoding algorithms, makes it possible to decode and reconstruct the sequence of brain activity associated with a cognitive task. Their drawback is that they cannot pinpoint the origin of the signals. Here too, methodological advances are underway, notably in very high-resolution EEG and in the emergence of a new generation of MEG sensors based on optical pumping and no longer requiring liquid helium cooling.

The lecture concluded with a brief overview of electrophysiological and optical imaging methods (2 and 3 photons) which are revolutionizing studies of the animal brain. These miniaturized techniques make it possible to record thousands of neurons simultaneously, or even all the neurons in an animal such as a zebrafish larva, with chronic monitoring of the same neurons day after day. The neural code, based on the collective activity of large populations of neurons, is beginning to become accessible, and we can, for example, decode precisely where an animal is, or what decision it is making, on the basis of the state of activity of neurons in the hippocampus or prefrontal cortex.