Semi-automated recognition of anatomical structures in flat maps of the cortical surface of the brain.

In the medical world there is still a strong demand for time-saving segmentation and labelling tools, in order to assist the neurosurgeon in the precise localisation of structures during stereotactic surgery or to aid the radiologist to better diagnose non-obvious localised lesions or to measure the volumes of lesions. On the long term, these systems might lead towards new views on brain anatomy, which is so far mainly based on thick sliced deformed cadaver brains. Systems, which could achieve the full anatomical labelling of an image within a few hours, give the anatomists the opportunity to create an anatomical model based on statistical values measured on hundreds of in-vivo brains, acquired with a MR scanner. At the same time the statistical knowledge of these models can be incorporated in the segmentation algorithms in order to further automate these methods, hence speeding up the whole labelling process in a way that it could be used to assist the diagnosis in the clinical routine. Within the frame of this project, we envisage to improve a particular medical labelling task: i.e. the labelling of the cortex of the brain because it can lead to better diagnosis and surgery, since cortical anatomy can hardly be understood from conventional 2D slice-views. It will be a semi-automatic approach based on the new combination of elastic matching, surface flattening and knowledge based labelling. The approach will be integrated in a versatile medical image analysis software environment, called TeDiMedIA software, developed at our departement in the context of previous regional and European projects.