%0 Journal Article
%T 基于自注意力增强DCGAN的肺结节影像样本生成技术研究
Research on Lung Nodule Image Sample Generation Based on Self-Attention Enhanced DCGAN
%A 任涵煜
%A 张致铭
%A 陈萌
%J Software Engineering and Applications
%P 267-273
%@ 2325-2278
%D 2026
%I Hans Publishing
%R 10.12677/sea.2026.152025
%X 基于肺结节医疗影像的自动化检测中训练样本不足的问题，重点探索了基于Attentional DCGAN (DCGAN-Attn)，一种在DCGAN架构中深度融合注意力机制的新型生成对抗网络的肺结节数据增扩技术。运用改良后的生成对抗网络技术对初始肺结节影像数据集实施数据扩充，并且在LIDC-IDRI数据集上予以验证。实验结果显示，与原始GAN相比，DCGAN-Attn技术在生成肺结节影像方面效果更好，FID指标均值达到87.19。并且通过多种生成图像质量评估指标综合测试，DCGAN-Attn网络生成的肺结节生成图像质量更高。
Addressing the issue of insufficient training samples in automated detection based on lung nodule medical images, this study focuses on a data augmentation approach using Attentional DCGAN (DCGAN-Attn), a novel generative adversarial network that deeply integrates an attention mechanism into the DCGAN architecture. The improved GAN technique is applied to augment the initial lung nodule image dataset, and validation is conducted on the LIDC-IDRI dataset. Experimental results demonstrate that, compared to the original GAN, the DCGAN-Attn method achieves superior performance in generating lung nodule images, with a mean FID score of 87.19. Furthermore, comprehensive evaluations using multiple image quality assessment metrics confirm that the lung nodule images generated by the DCGAN-Attn network exhibit higher quality.
%K 深度卷积生成对抗网络(DCGAN)，
%K 自注意力机制，
%K 医疗影像，
%K 数据增扩
Deep Convolutional Generative Adversarial Network (DCGAN)
%K Self-Attention Mechanism
%K Medical Imaging
%K Data Augmentation
%U http://www.hanspub.org/journal/PaperInformation.aspx?PaperID=140329