Some original slices of the IRM data.
First, I have registered both data sets to obtain one single 3D data volume, which is my head (still without the ears). For that task, I apply an original method of potential minimization (used to match multimodal medical images).
The segmentation of the head is quite easy. By a single thresholding, one can really easily remove the background in MR images. Some little connected components may remain (after thresholding) in the background, but they can be removed by considering only the greatest connected component in the thresholded image.
To obtain the 3D representation of the head, I used a surface rendering program.
The segmentation of the brain is more difficult. First, I had to find two thresholds which define the brain. The problem is that the brain is still connected to some other parts. To cut these connections, we use mathematical morphology. We apply an erosion to the result of the thresholding. Then we extract the greatest connected component of the eroded image. We obtain a kind of eroded brain. To reconstruct the brain, we apply a dilation to the eroded brain.
To obtain the 3D representation of the brain, I used a surface rendering program.
Having both segmentation (head and brain), I can put them together to obtain nice images, such the one below.
The 3D head was matched by Jacques Feldmar with an other 3D shape of my head obtained by stereoscopic reconstruction (with 2 cameras).
This allows either to project the texture of my face on the 3D shape coming from the MR images, or to project my brain onto one 2D image (it's called enhanced reality. Parental advertising: this image may cause offense to young people.).
From a 3-D MR image of the brain, we segment first the negative mould of the cortex, including part of the grey matter.
Second, we compute its skeleton.
The sulcal are the junction lines between the convex hull of the brain and the sulci.
