ABSTRACT Accurate synthesis of computed tomography (CT) images from magnetic resonance (MR) scans is essential for reducing radiation exposure and enabling fully MR‐based workflows in clinical applications such as radiotherapy planning. To address this task, we propose a novel framework—Pearson correlation coefficient‐guided dual‐attention network (PDANet)—that integrates dynamic supervision and attention‐driven representation learning for MR‐to‐CT image synthesis. PDANet tackles two key limitations in existing methods: fixed loss weighting and insufficient feature modeling. First, a Pearson correlation coefficient‐guided dynamic supervision strategy is introduced to adaptively balance pixel‐wise and perceptual losses throughout training, which allows the model to emphasize perceptual consistency in early stages and gradually shift toward pixel‐level refinement as structural similarity improves. Second, a dual‐attention mechanism is incorporated into the generator to enhance long‐range dependency modeling and spatially aware feature representation, improving the synthesis of anatomical structures across varying scales. Experimental results on two publicly available datasets—pelvis and brain—demonstrate that PDANet consistently outperforms state‐of‐the‐art methods in terms of structural fidelity and visual quality, highlighting the robustness and generalizability of the approach across diverse anatomical regions.
Zhu et al. (Thu,) studied this question.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: