HF-DRCT: A High-Frequency Enhanced Dense-Residual-Connected Transformer for Low-SNR 3D-SIM Reconstruction

3D Structured Illumination Microscopy (3D-SIM) enables high-resolution volumetric imaging of subcellular structures, but low signal-to-noise ratio (SNR) conditions can cause noise amplification, artifacts, and structural loss during reconstruction. Existing learning-based SIM methods often process axial sections independently, which limits their ability to use volumetric continuity. To address this limitation, we propose a High-Frequency Enhanced Dense-Residual-Connected Transformer (HF-DRCT) for low-SNR 3D-SIM reconstruction. HF-DRCT combines a multi-slice input strategy, a High-Frequency Enhancement Block, and a structure-aware composite loss to improve adjacent-slice context modeling, frequency-aware feature modulation, and structural detail preservation. High-SNR Wiener reconstructions are used as supervised reference labels rather than independently measured optical ground truth, so PSNR and SSIM indicate similarity to these references. Simulated experiments on multiple biological structures show that HF-DRCT achieves higher reference-based quantitative performance than Wiener deconvolution and existing deep learning baselines. Real low-SNR experiments further suggest that HF-DRCT suppresses background artifacts and preserves more coherent biological morphology in practical 3D-SIM reconstruction.