Enhanced reconstruction method for regions of interest in pet images based on multi-task learning constraints

Abstract

The invention discloses a PET image region-of-interest enhancement reconstruction method based on multi-task learning constraints. The method first obtains the back-projection image of PET original data in the image domain, and the main task of design and reconstruction is to use a three-dimensional deep convolutional neural network to establish a back-projection image. Mapping between projected images and PET reconstructed images. A new auxiliary task is designed to predict the computed tomography (CT) image with the same anatomical structure as the PET reconstruction image from the back-projection image, so that the local smoothness information of the high-resolution CT image can be used to reduce the noise in the PET reconstruction image. The new task 2 is designed to distinguish between the region of interest and the background region in the back-projection image, and to enhance the reconstruction of the region of interest during the reconstruction process, reduce the quantitative error caused by the smoothing of the region of interest, and improve the accuracy of PET reconstruction.

Description

technical field [0001] The invention belongs to the technical field of medical imaging, in particular to a method for enhancing and reconstructing a region of interest in a PET image based on multi-task constraints. Background technique [0002] Positron Emission Tomography (PET) is a medical image that can simultaneously provide information on human biological function metabolism and morphological anatomical structure. has been widely applied. The PET imaging process includes injecting a radioactive tracer into the patient before scanning; the tracer decays to produce positrons when it participates in physiological metabolism in the human body; the positrons have an annihilation effect with electrons in adjacent tissues, resulting in high-energy photon pairs moving in reverse ; Use two detectors with opposite positions to detect the photon pair to record a coincidence response line; collect a certain number of coincidence response lines, and arrange them into three-dimensi...

AI Technical Summary

Problems solved by technology

However, the current single-task-based deep learning reconstruction of PET images cannot introduce prior knowledge about PET images into the reconstruction map, and cannot introduce attention mechanisms to regions of interest in PET images, and the generalization ability is weak. Good models run the risk of producing artifacts when reconstructing new data

Therefore, deep learning-based reconstruction methods are currently not accepted by clinicians

Images