A method for transforming and segmenting medical volumes based on generating an antagonistic network

Abstract

A method for transforming and segmenting medical volumes based on generating an antagonistic network mainly comprises the acquisition of synthetic data, the interaction between a generator and a segmentation device, the complete target and the architecture of the antagonistic network and the process is that firstly, the generator is trained with the cyclic consistency loss function of the antagonistic network to compulsorily reconstruct the synthetic data so as to reduce the geometrical distortion of the synthetic image; then, two auxiliary maps, namely segmentation devices, are used to control the geometrical invariance of the synthesized data, so as to ensure the shape consistency of the synthesized image, and finally, the segmentation devices are amplified with the synthesized data generated in the generator to become the multi-mode segmentation devices to achieve efficient volume segmentation. Compared with the traditional method, the method can learn unpaired data, achieves the consistency of image set shapes and improves the segmentation efficiency by using synthetic data, thus achieving the goal of synthesizing clear 3D medical images efficiently.

Description

technical field [0001] The invention relates to the field of image processing, in particular to a method for transforming and segmenting medical volumes based on generating an adversarial network. Background technique [0002] With the development of computers and the emergence of digital instruments, researchers began to try to convert medical analog images into digital images, and carried out research on processing medical images with computers; medical image processing refers to comprehensive medical imaging, mathematical modeling, digital Interdisciplinary fields of image processing and analysis, artificial intelligence and numerical algorithms. In the process of diagnosis, the use of medical image processing technology can assist doctors to interpret medical images to a certain extent, eliminate human subjective factors, and improve the accuracy and efficiency of diagnosis; in the process of treatment, medical image processing technology can track the development of the...

AI Technical Summary

Problems solved by technology

[0004] Aiming at the problems that the existing medical image processing methods cannot learn unpaired data, the shape consistency of the synthesized image is low, and the efficiency of performing image segmentation tasks is not high, the purpose of the present invention is to provide a method for transforming and segmenting medical volumes based on generative adversarial networks The method of GAN first trains the generator with the cycle consistency loss function of the generative adversarial network to force the reconstruction of the synthetic data, thereby reducing the geometric distortion of the synthetic image; then, two auxiliary maps, namely the segmentor, are used to control the synthetic data The geometric invariance of the generator to ensure the shape consistency of the synthetic image of the generator; finally, the segmentor is augmented with the synthetic data generated in the generator, making it a multimodal segmenter for efficient volume segmentation

Images