Automatic segmentation process of 3D medical images by one or several neural networks through structured convolution according to the anatomical structure of 3D medical images

摘要

The present invention uses a knowledge database containing information on anatomical and pathological structures or instruments that can be seen in a 3D medical image consisting of n different 2D images of axbxn dimension, i.e., each of which is axb dimension. It relates to the segmentation method. This method creates 9 sub-images (1 to 9) of dimension a/2 xb/2 xn from the medical image, i.e., 9 partially overlapping a/2 xb/2 sub-images from each 2D image. A first step consisting of extracting; A second step consisting of analyzing and segmenting each sub-image among the nine sub-images (1 to 9) of each 2D image by nine convolutional neural networks (CNN); And the result of nine analysis and segmentation of n different 2D images, and thus the result of nine segmented sub-images of dimension a/2 xb/2 xn, corresponding to a single segmentation of the original medical image, It is characterized in that it mainly includes three process steps, which are a third step consisting of combining into a single image of axbxn dimension.