5. Classification stage with a dedicated convolutional CT-Casting-Net

Since the result of segmentation is a binary image containing all abrupt variations in contrast, including artifacts, it is necessary to continue processing with a classification step. To do this, the binary images obtained slice after slice are first stacked and assembled to form a binary volume, then a labeling step is performed to associate a number with each discontinuity.

The center of mass of each of these discontinuities is then used to cut a set of three 64-by-64 cross-sections along the three main directions in space (see figure 1 ). These three...