Application II : Segmentation of histological sections and fusion with MR images

In neurosurgery, localization of deep brain structures is a crucial issue. For instance, a new surgical treatment of Parkinson's disease consisting of deep brain stimulation has been recently developed. Up to now, the target has been localised using ventriculographic landmarks or MR images in stereotactic conditions. To determine with high accuracy the target localisation, the key issue appears to be the construction of a 3D atlas of the human basal ganglia, designed for further propagation onto the MR acquisition of a given patient. Our goal is thus to build such an atlas by fusing histological data with a 3D MR image of the same subject.

Figure 3: Supervised segmentation of the histological sections. Left: contours of the basal ganglia manually segmented section-by-section. Right: 3D view of the putamen contours superimposed with the associated histological section.

This requires three steps: first an automatic 2D realignment of the histological sections in order to obtain a three-dimensional block, then a manual segmentation of the basal ganglia performed by an anatomist, and finally a 3D registration between the reconstructed and segmented block and the MR image. The first and last steps are based on a robust registration algorithm (see ()), whereas the second step, performed within the yav++ platform, consists in a manual segmentation of the histological sections. segment allowing to perform a section-by-section segmentation and at the same time control, and eventually correct for, the 3-dimensional validity of the structures outlined on the sections (see Figure 3). A key feature of yav++ that is used during this slice by slice segmentation is the seamless synchronisation of a 2D and a 3D view of both the images and the slice contours. The user manually traces each contour in the 2D view while checking its 3D consistency in the 3D view (see Figure 3).

Figure 4: Affine registration of the histological block with an MR T1 image of the same subject. From left to right: MR and histological block are superimposed with increasing opacity factors (left column: MR:1, Histo.: 0; middle: MR:0.5, Histo.: 0.5; right: MR: 0.2, Histo.: 0.8).

To visually inspect any 3D image registration results, an image fusing tool is provided to superimpose 2 registered images with an increasing opacity factor and various colormaps (see Figure 4).

Figure 5: 3D visualisation of contours superimposed on 3-planes views of the histological block (left) and the registered MR volume (left).