基于Swin Transformer的MRI超分辨率重建

MRI Super-resolution Reconstruction Based on Swin Transformer

在医学图像分割、配准、影像组学以及计算机辅助诊断过程中,高分辨率的医学图像可以显著提高实验精度。但采集的图像切片厚度较大,对后续的图像分析带来了困难。旨在利用超分辨率重建将MRI头部厚层数据变为薄层数据。为提高超分辨率的精度,将图像分成小块,目标输出层为O_Block,输入层为I_Block。O_Block的大小为16×16像素。为了更精确地处理不同方向的组织纹理现象,I_Block要大于O_Block。在实验中,I_Block分别取32×32、48×48、64×64。实验过程中采用Swin Transformer来实现MRI图像超分辨率以获得头部MRI超分辨结果。实验结果表明,通过该实验可以较好地输入厚层图像来预测薄层图像。

In the process of medical image segmentation,alignment,image histology and computer-aided diagnosis,high-resolution medical images can significantly improve the experimental accuracy.However,the thickness of the acquired image slices is large,which brings difficulties for the subsequent image analysis.The aim of this paper is to use super-resolution reconstruction to change the thick layer data of MRI head into thin layer data.In order to be able to improve the accuracy of super-resolution,we divide the image into small blocks,the target output layer is O_Block,while the input layer is I_Block.the O_Block is in the size of 16×16 pixels.In order to more accurately handle the phenomenon of organizing texture in various different directions,I_Block is larger than O_Block.In the experiment,I_Block was taken as 32×32,48×48,and 64×64,respectively.We used Swin Transformer to realize the super-resolution of MRI images and obtain the super-resolution results of head MRI.The experimental results show that the thick layer image can be input better to predict the thin layer image by this experiment.