CycleGAN、ACGAN在人工智能医疗器械数据增广中的应用

Application of CycleGAN and ACGAN in Artificial Intelligence Medical Device Data Augmentation

目的探究人工智能医疗器械领域中使用循环生成对抗网络(Cycle-Consistent Generative Adversarial Networks,CycleGAN)和辅助分类生成对抗网络(Auxiliary Classification Generative Adversarial Network,ACGAN)进行数据增广的方法。方法使用CycleGAN和ACGAN分别生成干扰图像和特定领域数据,对图像增加不规律的变换,对原始图像数据进行数据加工或应用生成对抗网络生成该领域所需的图像数据。结果在医学影像数据集上评估了本文提出方法的性能,结果表明,CycleGAN和ACGAN可有效生成逼真的医学影像,从而用于训练机器学习模型。结论该方法解决了人工智能领域图像数据不足的问题,保证了模型对该数据的不可见性,使后期模型评估结果更准确。

Objective To explore the method of data augmentation using cycle-consistent generative adversarial networks(CycleGAN)and auxiliary classification generative adversarial network(ACGAN)in artificial intelligence medical devices.Methods The CycleGAN and ACGAN were used to generate interference images and specific domain data,respectively.Irregular transformations were applied to the images to augment them,and the original image data was processed or fed into generative adversarial networks to generate the required image data for that particular domain.Results The performance was evaluated on a medical imaging dataset,and the results showed that CycleGAN and ACGAN could effectively generate realistic medical images that could be used to train machine learning models.Conclusion This method can solve the problem of insufficient image data in the field of artificial intelligence,while ensuring the invisibility of the data to the model,making the later model evaluation results more accurate.