In neuroradiology and neurosurgery, the modern diagnostic and therapeutic procedures make an increasing use of computer-aided and image-guided tools. High quality imaging modalities like Computed Tomographies (CT) or Magnetic Resonance Imaging (MRI) provide a non-invasive view of the internal organs of the human body, such as the brain. Those three-dimensional (3D) image modalities are useful not only to detect pathologies, but also to prepare and realise a surgical operation with the help of sophisticated computer or robot systems. Several surgical procedures have been set up by this way. Among them, the determination of anatomical location of intracerebral sites in reference to an external device fixed to the patient's head has been termed stereotaxy. This technique is very largely used for reaching deep intracerebral targets (deep seated tumours, targets for movement disorder treatment, etc.).
The use of computer-assisted techniques to automatically and accurately locate regions of interest (targets) in a 3D image of an organ like a brain is a must in this context. Classical image segmentation techniques, basically relying on pixel (voxel in 3D) values, are obviously not sufficient to cope with the enormous complexity of the brain anatomy. The solution lies in the use of a priori information for image segmentation, i.e. in the inclusion of additional knowledge about a standard brain in the segmentation process. In this context, computerised tools have been developed to represent this prior information. Specifically, Computerised Brain Atlases (CBA) are recognised as important tools in this context.Those atlases consist in large database containing an anatomical (and sometimes functional) description of a standard human brain. They represent a real map of the brain, picturing the different areas with their names and locations. Because of the inter-individual variability of the brain shape, in order to use this large prior information in a diagnostic or surgical procedure, for instance to automatically define target regions, those atlases have to be deformed to match the patient's brain anatomy. This is called registration. When the atlas is registered with the patient's images, his/her whole brain is entirely mapped.Depending of the medical application, this important structural prior information may then be used for instance for planning a neurosurgical operation or for quantifying functional MRI (fMRI) activation in specific regions of interest.
While many multimodal registration algorithms have been proposed in the past ten years, several issues still remain open. Specifically, based on our experience, we consider that it is important to involve both global and local deformation to achieve a very accurate registration. The main objective of this project is to develop non-rigid registration algorithm that benefit from the robustness of global non-rigid registration but that further improve the registration accuracy by considering local deformations.
The development of two algorithms has been initiated in the previous project and will be further developed in this project. The first one is based on the optimal placement of local control points to locally govern the deformation. The second one is based on local segmentation by active contours and on the propagation of the resulting displacement field to the neighbourhood of the contours by level-set tracking. Those algorithms will be applied both to segment brain structures in normal brain images for functional stereotactic procedures and to the segmentation of cerebral structures in brains containing large space-occupying lesions like tumours.
We will also continue our extensive study on the validation of non-rigid registration algorithm for atlas-based localisation of the Subthalamic Nucleus (STN) in MR images. We will include more algorithms in the study, as well as a ground truth coming from cadaver brains included in our Lausanne Brain Atlas. Finally, we will open the door for new investigations on non-rigid registration of multimodal atlases with multimodal images, i.e. the registration of atlases and data sets made not of one single image but of several modalities, such as T1-weighted or T2-weighted MR images, Diffusion-Tensor MR images, etc.
With this project, we will be able to automatically and accurately register a reference brain atlas with the patient's 3D images, involving validated algorithms. We will use those results in open and stereotactic neurosurgical procedures, in order to define brain targets and regions of interest with more accuracy, providing optimization of surgical treatment efficiency and minimizing adverse effects.