Multi-channel spatial attention network for MRI super-resolution

Magnetic resonance imaging (MRI) provides important structural and functional information for clinical diagnosis. Due to the limitations in the device or the scanning time, obtainment of high-resolution MRI is not always available. MRI super-resolution techniques achieve high-resolution reconstruction without prolonging scanning time, reduce the interference of motion artifacts on image quality, and provide a reliable basis for the precise detection and quantitative analysis of minor lesions. We design a multichannel spatial attention network (MSANet) for MRI super-resolution. The core innovation of this network lies in the multi-channel spatial attention block (MSAB) which combines the global receptive block (GRB) with high-resolution spatial attention (HSA), enabling efficient capture of spatially continuous information. Meanwhile, the image decomposition path with discrete wavelet transform (DWT) further enriches the detail features. The MSANet is tested on brain, heart, and spine MRI datasets under 3x/4 × super-resolution reconstruction, which achieves PSNRs of 31.86 dB/30.23 dB, 37.12 dB/34.14 dB, and 34.48 dB/32.02 dB, respectively, outperforming state-of-the-art methods. In addition, this method also has a good effect on increasing the interlayer continuity of 3D MRI.