Vascular reconstruction method of MRI brain images based on 3t and 7t

Abstract

The present invention provides a 3T, 7T-based nuclear magnetic resonance brain image blood vessel reconstruction method: comprising the following steps: Step 1: Acquire 3T pictures and 7T pictures, perform image preprocessing on the 3T pictures and 7T pictures, and obtain the preprocessed 3T pictures pictures and 7T pictures; Step 2: Based on the U‑net network, use the preprocessed 3T pictures as the input of the U‑net network to obtain the output pictures through U‑net; Step 3: Separate the output pictures through U‑net Input the preprocessed 7T picture into the VGG-16 network to obtain the output of the 3T picture through the VGG-16 network and the output of the 7T picture through the VGG-16 network; Step 4: Output to the U-net network, and the 3T picture through the VGG The output after the ‑16 network and the output of the 7T image after passing through the VGG‑16 network are both calculated for loss. Based on the loss function, the parameters are obtained by using the stochastic gradient descent method, and the U‑net network is updated according to the parameters, and the 3T image is input into the updated U‑net The network gets reconstructed results. The invention can reconstruct blood vessels on 3T pictures.

Description

technical field [0001] The invention relates to the technical field of magnetic resonance imaging, in particular to a 3T and 7T-based nuclear magnetic resonance brain image blood vessel reconstruction method. Background technique [0002] 7T MRI images can clearly see blood vessels that 3T MRI images do not have, but due to the high cost and scarcity of 7T equipment, most hospitals still use 3T images. Therefore, aiming at this problem, it is very necessary to propose a vascular reconstruction method based on 3T and 7T MRI brain images to overcome this problem. [0003] Most of the existing domestic and foreign researches on image reconstruction focus on super-resolution image reconstruction. In 2014, Dong et al. proposed a CNN model SRCNN for general natural image super-resolution reconstruction. Based on SRCNN, Kim et al. proposed VDSR by referring to the VGG network structure for image classification. Bee Lim et al. proposed an enhanced deep residual network EDSR. [...

AI Technical Summary

Problems solved by technology

[0002] 7T magnetic resonance images can clearly see blood vessels that 3T magnetic resonance images do not have, but due to the high cost and scarcity of 7T equipment, most hospitals still use 3T images

[0007] (2) Unlike under-sampled low-resolution images and high-resolution images, 3T and 7T images are not completely registered, and it is difficult to reconstruct images directly using the differences between images

Images