A unified deep network for thin- and dense-slice reconstruction: Improving through-plane resolution in clinical MRI

Magnetic Resonance Imaging (MRI) is widely used in clinical diagnostics for its high soft-tissue contrast and non-invasive nature. However, to accelerate acquisition and reduce patient burden, clinical protocols often trade off through-plane resolution, leading to either sparse inter-slice sampling or thick-slice acquisition, which impairs 3D reconstruction and anatomical interpretation. This work aims to develop a unified super-resolution framework for enhancing Z-axis resolution in MRI, capable of handling both Sparse-to-Dense and Thick-to-Thin scenarios commonly encountered in clinical practice. We propose an end-to-end 3D MRI super-resolution network that combines task-specific pre-processing with a shared architecture. The model incorporates enhanced residual blocks with spatial and channel attention, a slice attention module to capture inter-slice dependencies, and a 3D PixelShuffle-based upsampling mechanism for anisotropic resolution recovery. Extensive experiments on public brain MRI datasets demonstrate that the proposed method consistently outperforms existing baselines across multiple scale factors and degradation settings. It achieves superior PSNR, SSIM, and MSE scores while preserving fine anatomical details and volumetric consistency. The proposed framework provides an effective and generalizable solution for MRI Z-axis resolution enhancement, bridging clinically relevant acquisition gaps and supporting accurate 3D reconstruction under diverse degradation conditions.