Analysis and Processing of Anatomical Brain MR Images

Magnetic resonance imaging (MRI) is one of the most frequently used modalities in medical imaging, alongside X-ray and ultrasound imaging. MRI combines the advantages of these latter modalities, without suffering from their weaknesses. It provides a high level of spatial and spectral resolution, without inducing radiation harmful to patients, or requiring (in most cases) the injection of contrast agents.

Thanks to these qualities, MRI has become the preferred image acquisition modality for most medical examinations related to brain disorders. In this context, specific acquisition sequences have been progressively developed to meet the needs of particular anatomical or pathological structures (vascular networks, tumors, etc.) for patient diagnosis, follow-up or treatment. MRI scanners are now standard equipment in hospitals.

For the same reasons of efficiency and safety, MRI is a remarkable research tool for studying the human brain in vivo. Here too, specific acquisition sequences enable progress to be made in understanding the structure of the brain, its development (cerebral maturation in the foetus and young child) and its evolution (degeneration linked to ageing), as well as its physiological and cognitive functioning.

However, users of MRI images face a number of challenges related to the nature and content of these images. The first difficulty stems from the constant progress made by MRI scanner manufacturers. Two-dimensional images have now given way to 3D and even 4D images (images in space and time). Volumes of information are now so great that analysis by the human eye alone is no longer possible. Image resolution is also increasing, now reaching sub-millimetre values. This finesse of detail, combined with the very high anatomical complexity of the human brain, leads to a second difficulty, linked to the semantic analysis of MRI images, which – if based on human expertise – can no longer do without computer assistance.

In this context, image processing and analysis approaches are being developed in order to produce software tools capable of assisting medical experts and researchers in their use of MRI images. In particular, a wide range of issues will be addressed, from signal processing to image semantics. First and foremost, MRI images need to be made easier to read, by removing as far as possible any visual artifacts due to the physical conditions of their acquisition (§ 1