A nuclear magnetic resonance multi-weighted imaging method based on a depth generative adversarial neural network

Abstract

The invention relates to a nuclear magnetic resonance multi-weighted imaging method based on a depth generative adversarial neural network, which comprises the following four steps: the constructionof the depth generative adversarial neural network, the construction of a training data set and an evaluation data set, the training of network weights and the application of the network weights to aplurality of weighted nuclear magnetic resonance imaging. More than 40,000 pairs and 10,000 pairs of T2-weighted and PD-weighted NMR images collected from the same site at the same time were used as training data set and evaluation data set respectively, The model weight of the depth generative adversarial neural network is trained, and the generation network uses the T2-weighted image as the input of the network, and maps the data distribution of the generated image to the PD-weighted image data distribution obtained by acquisition maximally; The invention can convert the T2-weighted nuclearmagnetic resonance image acquired by the nuclear magnetic resonance imaging device into a high-quality PD-weighted nuclear magnetic resonance image in a very short time, thereby providing two kinds of weighted nuclear magnetic resonance images in a single imaging process.

Description

technical field [0001] The invention belongs to the field of medical nuclear magnetic resonance imaging, and in particular relates to a nuclear magnetic resonance multi-weighted imaging method based on deep generative adversarial neural networks. Background technique [0002] In the existing traditional single-weighted magnetic resonance imaging technology, the signal values ​​of different tissue parts received by the magnetic resonance imaging equipment are as follows: [0003] M=M 0 (1-e -TR / T1 )e -TE / T2 [0004] where M 0 is the magnetization vector, TR is the interval time between two consecutive pulses, TE is the time between the 90° pulse and the spin echo, T1 is the time required for the longitudinal magnetization to recover to 63% of the original magnetic vector, and T2 is the time for the transverse magnetization to decrease The time required to be as small as 37% of the original magnetic vector. The traditional single-weighted MRI method uses different TR and...

AI Technical Summary

Problems solved by technology

[0009] Aiming at the defects and deficiencies of the traditional single-weighted nuclear magnetic resonance imaging technology, the present invention provides a multi-weighted nuclear magnetic resonance imaging method based on deep generative adversarial neural networks, so as to provide two kinds of weighted nuclear magnetic resonance images in one imaging process

Images