Low-dose PET image reconstruction algorithm based on ADMM and deep learning

Abstract

The invention discloses a low-dose PET image reconstruction algorithm based on ADMM (Amplitude Division Multiplexing Model) and deep learning, which solves a maximum likelihood reconstruction model into three sub-problems, namely a reconstruction layer, a denoising layer and a multiplier layer, nests an iterative reconstruction framework, and optimizes and reconstructs low-dose PET projection data by utilizing a deep learning thought. The reconstruction layer uses a traditional EM reconstruction kernel, the denoising layer uses a residual convolutional neural network for representation, the neural network is embedded into a traditional iterative reconstruction framework, reconstruction and training are realized at the same time, and a high-quality low-dose reconstructed image is obtained. According to the method, the traditional reconstruction and the neural network are successfully combined, and the problems of end-to-end learning lack of a reconstruction kernel and low traditional iteration speed of the neural network are solved.

Description

technical field [0001] The invention belongs to the technical field of biomedical image analysis, and in particular relates to a low-dose PET image reconstruction algorithm based on ADMM and deep learning. Background technique [0002] Positron Emission Tomography (PET) is an in vivo molecular imaging technique widely used clinically, especially in the screening process of cancer; PET can sensitively measure radiotracer The distribution of the human body, and reconstruct this distribution through different methods. In recent years, many studies have focused on the radiation problems caused by radiotracers, so low-dose PET reconstruction has received attention. However, the problem faced by low-dose PET reconstruction is the reduction of image quality due to the reduction of data volume, that is, the reduction of image Signal-to-Noise Ratio (SNR). In order to balance the relationship between scanning dose and image quality, many technologies have been proposed, such as enha...

AI Technical Summary

Problems solved by technology

[0005] In these deep learning-based end-to-end training processes, the training data is often processed into images with a consistent noise level, but in actual PET scans, the noise level of the final data varies from person to person even with the same dose; and In traditional iterative reconstruction, the noise level of the input image is also changed for each iteration

Therefore, this brings challenges to the selection of training data in deep learning based on CNNs; all in all, it is imminent to develop a reconstruction kernel optimization algorithm with fast convergence speed and Poisson distribution characteristics combined with PET data

Images