PROCESS OF AUTOMATIC SEGMENTATION OF THREE-DIMENSIONAL MEDICAL IMAGE, USING ONE OR SEVERAL NEURAL NETWORKS, BY MEANS OF STRUCTURED CONVOLUTION IN ACCORDANCE WITH ANATOMICAL GEOMETRY OF THREE-DIMENSIONAL MEDICAL IMAGE Russian patent published in 2022 - IPC G06T7/11 G06T1/40 

FIELD: image processing.

SUBSTANCE: invention relates to devices and methods for image processing. Nine partially overlapped sub-images are extracted from a medical image; each of nine sub-images is analyzed and segmented, using nine convolutional neural networks; results of analysis and segmentation of n different two-dimensional images are combined into a single image corresponding to unified segmentation of the initial medical image.

EFFECT: increase in the accuracy of segmentation of a medical image.

6 cl, 10 dwg