脑皮层的立体脑回图展成平面的新方法

Flattening of the 3D Cortical Surface for Improved 2D Visualization

用核磁共振 (MRI)大脑断层图 ,分析患者的脑皮层解剖特征是一项困难的工作 .一个可能的途径是将分割后的脑皮层进行三维重建 .但是重建后的三维立体脑皮层图 ,如不经过 36 0°旋转仅能看到脑皮层上脑回的一部分 .因此还需要展开成平面脑回图以看到脑回的完整走向 .文中提出的投影方法能够在变形尽量小的前提下将立体脑回图展开在平面上 ,直观地看到脑回走向的全貌 。

Understanding cortical brain anatomy is a complex task, and quite impossible when consulting a MRI study of the brain in a slice by slice fashion. A better interpretation can be obtained via a 3D volume rendering of the segmented cortex. However classical volume rendering provides only a partial view of the cortical surface, from a certain angle. Therefore several renderings have to be made, and interpreted all together to identify the sulci of the brain. In this paper, a projection method is presented. First, a MR image set of the brain is segmented, i.e. all grey and white matter tissues are isolated via an approach based on the principles of seeded region growing, supervised clustering and morphological filtering. The spinal cord is cut at the level of the bottom of the cerebellum. Next, two hemi ellipsoids are shaped interactively by adjusting their principle axes and common center point, so that they fit as close as possible to the top and bottom parts of the brain. This interactive procedure is supported by a software, which provides three orthogonal views. The axes of the ellipsoids are defined on the orthogonal sections of maximal brain extent. For each hemi ellipsoid, a rectangular regular grid of points in the (u,v) plane of projection is attributed corresponding points on the ellipsoidal surface, so that the center of the original rectangle is mapped to the top of the hemi ellipsoid and the points lying on the perimeter of the grid are moved towards the perimeter at the bottom of the hemi ellipsoid. This wrapping yields points on the hemi ellipsoidal surface that are no longer uniform. In order to obtain good projection results, the distances are homogenized by an iterative method. Once the points are homogenized, both the hemi ellipsoidal surfaces are mapped towards the binary image of the brain by means of a deformable surface model. After each deformation step, the distances between the points are homogenized by an enhanced version of the Fixed Point algorithm. At last, a ray is fired for each point normal to the obtained dilated surface. At the intersection of these rays with the original segmented brain, we compute the shading analogous to volume rendering. The projection method described in this paper has the advantage that the complete cortical surface is unwrapped in a 2D plane without severe geometric distortion. Sulci can be followed in their entirety, so that it is much easier to recognize them.