Focus detection model training method based on generative adversarial network

Abstract

The invention discloses a focus detection model training method based on a generative adversarial network, a device, equipment and a readable storage medium. According to the method, two ideas are adopted to reduce the requirements of the generative adversarial network for data: on one hand, a normal image, a focus and a focus mask are synthesized into three-channel image data to serve as the input of the network, and the three channels respectively contain different prior information, so that the difficulty of image generation is reduced; and on the other hand, a local evaluation mode is adopted in the evaluator of the generative adversarial network, and the requirement on the depth of the network is reduced in the mode, so that the requirement on the sample size is also reduced. Finally,the method achieves the purpose of completing the training of the generative adversarial network by using the small sample, thereby effectively expanding the training sample of the focus detection model, and facilitating the improvement of the detection performance of the focus detection model.

Description

technical field [0001] The present application relates to the field of computer technology, and in particular to a training method, device, equipment and readable storage medium of a lesion detection model based on a generative adversarial network. Background technique [0002] Worldwide, cancer remains the greatest threat to human life. Taking breast cancer as an example, breast cancer has always been the cancer with the highest incidence rate in women. Early treatment of breast cancer can not only prolong the life of patients, but also greatly reduce the overall expenditure of patients in terms of treatment costs. Therefore, artificial intelligence-based mammography-assisted diagnosis of breast cancer has gained more and more attention from doctors. [0003] In recent years, with the success of deep learning in natural images, people are also trying deep learning-based methods for cancer diagnosis and screening. However, deep learning methods based on convolutional neura...

AI Technical Summary

Problems solved by technology

[0007] The purpose of this application is to provide a training method, device, equipment, and readable storage medium for a lesion detection model based on a generative confrontation network, so as to solve the problem that the traditional sample expansion scheme based on a generative confrontation network requires more samples to complete pair generation. The training of the adversarial network leads to the problem that the sample expansion effect of the adversarial network is not ideal in the case of small samples, and it is impossible to improve the performance of lesion detection

Images