Heart magnetic resonance image data enhancement method based on evolutionary GAN

Abstract

The invention relates to a heart magnetic resonance image data enhancement method based on evolutionary GAN. According to the method, when a generator is trained, the generator is mutated to generatea plurality of sub-generation generators; the adaptability scores of the generators are judged through an adaptability score function; an optimal filial generation generator is selected as a parent generation generator of the next iteration according to the score; meanwhile, in the discriminator training stage, a new training sample is synthesized in combination with linear interpolation of the feature vector, and a related linear interpolation label is generated, so that the distribution of the whole training set is expanded, the discrete sample space is continued, the inter-domain smoothnessis improved, and the model can be better trained. According to the image enhancement method, high-quality and diverse samples can be generated to expand the training set, and finally various indexesof the classification result are improved.

Description

technical field [0001] The invention relates to the field of image processing, in particular to a method for enhancing cardiac magnetic resonance image data based on evolutionary GAN. Background technique [0002] Cardiac magnetic resonance is known as the gold standard for assessing cardiac function. Conventional cardiac magnetic resonance scanning technology has been relatively mature and has played a vital role in disease diagnosis. At present, many cardiac magnetic resonance image-aided diagnosis tasks based on deep learning have achieved good results, but cardiac magnetic resonance images not only require expensive medical equipment to obtain, but also require a large amount of manual data annotation by experienced radiologists , which is undoubtedly extremely time-consuming and labor-intensive. In addition, the privacy of patients in the field of medical images has always been quite sensitive, so obtaining a large number of data sets with balanced positive and negativ...

AI Technical Summary

Problems solved by technology

At present, many cardiac magnetic resonance image-aided diagnosis tasks based on deep learning have achieved good results, but cardiac magnetic resonance images not only require expensive medical equipment to obtain, but also require a large amount of manual data annotation by experienced radiologists , which is undoubtedly extremely time-consuming and labor-intensive

In addition, the privacy of patients in the field of medical images has always been quite sensitive, so obtaining a large number of positive and negative sample balance data sets requires a very large cost

On the one hand, GAN is very difficult to train. Once the data distribution and the distribution fitted by the generated network do not substantially overlap at the beginning of the training, the gradient of the generated network will easily point to a random direction, resulting in the problem of gradient disappearance.

On the other hand, the generator will try to generate a single sample that is safer but lacks diversity in order to allow the discriminator to give a high score, which will lead to the problem of mode collapse

Images