Medical image super-resolution reconstruction based on generative adversarial networks (GAN)

Medical imaging super-resolution technology is critical for improving diagnostic accuracy, yet traditional methods face challenges such as blurring artifacts and detail loss, while existing deep learning models risk distorting pathological features. This study proposes a Dual-Path Progressive Generative Adversarial Network (Dual-ProGAN), leveraging GANs' high-fidelity generation. The framework employs a dual-path mechanism: the low-frequency path utilizes dense residual blocks to recover organ morphology and lesion localization, while the high-frequency path enhances sub-millimeter pathological details (e.g., microcalcifications) via dynamic attention. A gated fusion strategy dynamically balances anatomical and functional features. Training adopts progressive resolution enhancement with multi-scale discriminators and gradient-sensitive loss functions to optimize pixel accuracy, structural similarity, and pathological consistency. This solution advances medical imaging super-resolution, supporting primary healthcare equipment upgrades and precision surgical planning.