【摘要】 磁共振成像(Magnetic Resonance Imaging,MRI)是一种重要的医学成像技术。图像超分辨率重建能够在缩短扫描时间的同时提升图像清晰度和细节表现,为临床诊断提供更高质量的影像支持。针对MRI图像超分辨率重建任务中高频细节和多尺度信息恢复不足的问题,本文提出了一种结合频域约束、多尺度特征提取与扩散重建的图像超分辨率重建方法,并在MRI图像上进行了实验验证。该方法首先引入傅里叶变换损失和小波变换损失,以增强模型对频域特征和多尺度细节信息的建模能力;随后利用基于Residual Swin Transformer Block(RSTB)的特征提取模块来提取残差特征,提升了对高频信息和结构特征的表达能力;最后结合扩散模型实现高质量图像重建。在IXI数据集和自采临床癫痫T2加权(T2-weighted,T2w)MRI数据集上的实验结果表明,该方法在峰值信噪比、结构相似性指数和感知图像块相似度等指标上均表现较优,能够较好地恢复图像细节,同时能保持图像整体结构与纹理信息的一致性。更多还原

【Abstract】 Magnetic resonance imaging(MRI) is an important medical imaging modality,and image super-resolution can improve image clarity and detail while reducing scan time for clinical diagnosis.To address the insufficient recovery of high-frequency details and multi-scale information in MRI image super-resolution reconstruction,this paper proposes a method integrating frequency-domain constraints,multi-scale feature extraction,and diffusion-based reconstruction.Fourier transform loss and wavelet transform loss are introduced to enhance frequency-domain and multi-scale detail modeling,while a residual feature extraction module based on the Residual Swin Transformer Block(RSTB) is employed to improve the representation of high-frequency and structural information.Experimental results on the IXI dataset and a self-collected clinical epilepsy T2-weighted(T2w) MRI dataset demonstrate that the proposed method achieves competitive performance in terms of peak signal-to-noise ratio,structural similarity index measure,and learned perceptual image patch similarity,while effectively preserving image details,structural consistency,and texture information.更多还原