JoWUA

Abstract

Medical image generation plays an important role in modern healthcare systems, addressing data deficiencies to improve the effectiveness of diagnostic models. GANs have shown impressive potential for producing high-quality synthetic images; however, they often fail to maintain structural consistency and preserve fine image details. To overcome these challenges, this study proposes a novel hybrid attention-driven generative adversarial network that incorporates spatial and channel attention, as well as multi-scale feature fusion. Attention mechanisms have shown promising potential in producing high-quality images that are more interpretable and meaningful in a clinical decision-making process. The proposed model is implemented and tested using various datasets and evaluation metrics, including PSNR, SSIM, and FID. Experimental results show that HA-GAN achieves superior performance compared to existing models in producing high-quality images for decision-making. Image generation in medical diagnostics deals with the problem of data insufficiency in diagnostic techniques. GANs encounter issues with the structural accuracy of their generated images. The authors of this study have used attention-driven Hybrid GAN (HybridAGAN) with multi-scale attention features to resolve the above-mentioned problems. Specifically, the model's generator combines an attention mechanism with a multi-scale representation. At the same time, the discriminators focus on both the global structure and the local texture of generated images. The effectiveness of the proposed approach is validated using two benchmarks: the Brain MRI and Chest X-ray datasets. To measure the model's performance, several quantitative criteria were used, including PSNR, SSIM, and FID. The results showed that the proposed model performed much better than the baselines, achieving 34.7 dB PSNR, 0.91 SSIM, and 21.4 FID. Statistical testing, including paired t-tests (p < 0.05), demonstrated the significance of the obtained results.