Focus image generation method and device, electronic equipment and storage medium

Abstract

The invention discloses a lesion image generation method and device, electronic equipment and a storage medium, and the method comprises the steps: carrying out the image recognition processing of each historical lesion image in at least one historical lesion image, and obtaining at least one mask image in one-to-one correspondence with the at least one historical lesion image; performing superposition processing on each mask image and the corresponding historical focus image to obtain at least one superposition image in one-to-one correspondence with the at least one mask image; training an improved cyclic adversarial neural network according to the at least one superimposed image, the at least one mask image and the at least one historical lesion image to obtain a lesion image generation network; acquiring an original image of a to-be-generated focus; randomly selecting one or more images from the at least one mask image, and carrying out superposition processing on the selected one or more images and the original image to obtain a to-be-processed image; and inputting the to-be-processed image into the focus image generation network to generate a focus image.

Description

technical field [0001] The present invention relates to the technical field of artificial intelligence, in particular to a lesion image generation method, device, electronic equipment and storage medium. Background technique [0002] With the gradual digitization of imaging equipment and the development of computer communication technology, digital image transmission and electronic film came into being. Many traditional hospitals have transformed into digital hospitals one after another. This not only speeds up the efficiency of outpatient consultation, but also increases doctors' attention to medical imaging data. Audit volume. Therefore, using deep learning technology to realize intelligent diagnosis and assistance of medical images to reduce the workload of doctors has become a current research hotspot. Generally speaking, deep learning technology realizes the training of the model by learning the characteristics of the data independently through the network. The perform...

AI Technical Summary

Problems solved by technology

Taking brain CT data as an example, although brain CT data grows at a certain rate every year, because cerebral hemorrhage is not a very common disease, and there are many types of lesions, it is difficult to collect brain CT data with lesions. Most of the data are normal CT data

Even if a certain amount of samples with lesions can be collected in the end, due to the diversity of lesions, the data volume of various types of lesions is seriously unbalanced. At the same time, medical data can only be marked by professional doctors, which takes a lot of time, manpower and material resources cost, which greatly affects the development cycle of the model

[0004] In view of the above-mentioned problems of difficult collection of training samples, insufficient labeling, and high cost, the existing technical solutions mainly include the following: The first is to use commonly used image amplification techniques such as rotation, translation, and cropping to increase the sample size. , but this method is only sample expansion based on existing data, lacks diversity, and has little gain for the model; the second is image clipping technology, which adds the lesion area to another image by cutting and pasting it, although This method can make up for the lack of diversity in method 1, but because the pasted lesions are relatively blunt in excessive texture, too many such samples are likely to mislead network learning and affect performance; the third is to use the confrontation generation network (Generative Adversarial Networks, GAN) for sample expansion, this method can not only generate a large number of samples, but also solve the problem of the diversity of generated samples and whether the samples are realistic to a certain extent

However, if the existing GAN technology wants to generate a large number of realistic samples, it also requires a large number of training samples. At the same time, the location of the lesion in the generated image is uncontrollable, and it is easy to generate data that is inconsistent with the mechanism of the actual brain disease.

Images