PET quick imaging method and system based on deep learning

Abstract

The invention discloses a PET quick imaging method and system based on deep learning. The PET quick imaging method comprises the steps of obtaining the matched nuclear medicine image sets collected ina first long time and a second short time; preprocessing the nuclear medicine image group and normalizing the image size and the signal intensity; constructing a noise recognition neural network based on Res-UNet, wherein the noise recognition network uses a network output image composed of a network input image formed by randomly adding a noise signal to the acquired nuclear medical image and the noise signal; constructing an image reconstruction convolutional network which is provided with a plurality of convolutional neural network CNN levels connected in series and comprises an encoder-decoder residual depth network structure which is symmetrically connected in series, wherein the input is an image acquired in a second short time and a multi-contrast image used as multi-modal input; and training the image reconstruction convolutional network to enable the image reconstruction convolutional network to priori extract effective image features of the input nuclear medical image and the multi-contrast image so as to reconstruct a high-quality nuclear medical image acquired for a long time.

Description

technical field [0001] The present invention relates to PET image reconstruction technology, especially a deep learning-based PET rapid imaging method and system. Background technique [0002] Positron Emission Computed Tomography (PET) is a medical imaging technology that injects positron radioactive isotope-labeled compounds into organisms and collects decay positron signals for imaging during biological metabolism. Since the time required to collect positron signals depends on the half-life of radioactive elements and the effective time interval of the signal acquisition system, and the traditional imaging algorithm has the characteristics of a large amount of calculation, the PET imaging faces the problem of long imaging time. In order to improve the efficiency of PET imaging, people try to reduce the data acquisition time to acquire images. However, the method of shortening the imaging duration by shortening the signal acquisition time faces problems such as low signal...

AI Technical Summary

Problems solved by technology

[0007] The main purpose of the present invention is to provide a PET rapid imaging method based on deep learning, which solves the problems in the prior art such as shortening the imaging duration and causing image quality degradation.

Images