Medical Image Super-Resolution Reconstruction Model DGS-SRGAN based on Dual-Granularity Synergistic Strategy

Lung adenocarcinoma is a malignant tumor that originates in the lung gland cells and accounts for more than 50% of all lung cancer. It is often asymptomatic in its early stages, resulting in patients being diagnosed at a later stage and missing the best time to treat it. Regular CT scan is an effective means to realize early diagnosis and treatment of lung adenocarcinoma. However, low-dose CT images are often affected by noise interference and detail loss, affecting the accuracy of diagnosis. To solve this problem, this paper proposes a super-resolution reconstruction model (DGS-SRGAN) based on two-granularity synergistic strategy. The model optimizes the generator's characteristic receptive field through deep Macrokernel attention (LKA), integrates the Dual-granularity Synergistic attention (DGSA) module, integrates multi-scale spatial attention and progressive channel self-attention mechanism, and thus significantly improves the detail reconstruction ability. In the discriminator design, the two-branch cooperative strategy of global context awareness and local detail analysis enhances the discriminant ability of image authenticity, and introduces an adaptive dynamic adaptation mechanism to process multi-scale and multi-granularity image input. Through ablation experiment, comparison experiment and cross-modal verification, the performance of DGS-SRGAN model in medical image super-resolution reconstruction was comprehensively evaluated. The experimental results show that in the super-resolution tasks of lung CT images, the PSNR of the 2x and 4x super-resolution tasks are improved by about 3.1dB and 2.0dB, respectively, and the SSIM is improved by about 5.4% and 3.2%.