TRCA-GAN: Tumor Representation Consistency Alignment for Cross-Subject Multimodal MRI Synthesis
Can Chang, Li Yao, Xiaojie Zhao · 2025
Multimodal MRI synthesis is crucial for clinical diagnosis and treatment planning, particularly in glioma segmentation. However, cross-subject synthesis faces significant challenges due to the variability in brain structures and tumor morphology across individuals. Existing methods often fail to preserve tumor region fidelity, limiting their practical utility. To address these limitations, we propose the Tumor Representation Consistency Alignment Generative Adversarial Network (TRCA-GAN), a novel framework designed to enhance tumor region fidelity in cross-subject multimodal MRI synthesis. TRCA-GAN introduces two key innovations: the Region-Aware Perturbation Module (RAPM), which enhances tumor structure preservation through adaptive perturbation and joint attention mechanisms, and the Local Contrast Discriminator Module (LCDM), which improves tumor contrast fidelity via a dual physical-feature constraint mechanism. Extensive experiments on the BraTS2021 dataset demonstrate that TRCA-GAN significantly outperforms comparison methods in terms of image generation quality and tumor segmentation accuracy. The results highlight the effectiveness of TRCA-GAN in addressing the challenges of cross-subject MRI synthesis, making it a powerful tool for clinical applications.