Deep learning method for medical image low-dose estimation

Abstract

The invention discloses a deep learning method for medical image low-dose estimation. The method comprises the steps that wavelet transform is used for decomposing a low-dose original image to obtain multiple layers of decomposed images, and each layer of decomposed image comprises multiple sub-band images with different view angle characteristics corresponding to the original image; the multiple layers of decomposed images and the original image are input into a convolutional neural network for training, a mapping relation from a low-dose original image to a standard-dose image is learned through decomposition and reconstruction, the convolutional neural network comprises a trunk structure and a plurality of branch structures, the trunk structure takes the original image as input, and the plurality of branch structures respectively take the corresponding decomposed images of each layer as input. According to the method, the wavelet transform is combined with the convolutional neural network, so that the image detail information is enhanced while the image peak signal-to-noise ratio, the structural similarity and the contrast signal-to-noise ratio are improved.

Description

technical field [0001] The present invention relates to the technical field of medical image processing, and more specifically, to a deep learning method for low-dose estimation of medical images. Background technique [0002] Medical image processing is widely used in clinical guidance. For example, positron emission tomography-magnetic resonance imaging (PET-MRI) is a hybrid imaging technique that combines soft tissue morphological imaging with magnetic resonance imaging (MRI) and functional imaging with positron emission tomography (PET). Compared with other methods, PET-MRI examination has low radiation dose, high sensitivity, and good accuracy, and has the value of early detection and early diagnosis of many diseases (especially tumors and the most common heart and brain diseases). However, most of the radioactive radiation used in PET comes from fluorine-18 (18F), which emits positrons and produces high-energy gamma rays, which have certain radiation to the human body...

AI Technical Summary

Problems solved by technology

[0005] However, when low-dose radionuclides are used in PET-MRI, PET images will not only have a lot of noise, but also structural deformation, and traditional PET low-dose denoising techniques are usually based on convolutional neural networks or traditional mathematical methods, poor image reconstruction

Images