A dynamic reconstruction method for enhancing the details of a pet image

By adding an attention SE module and improving the loss function to FBP-Net, the problem of balancing speed and quality in PET image reconstruction was solved, achieving high-quality, high-detail PET image reconstruction and improving diagnostic accuracy.

CN116309898BActive Publication Date: 2026-07-03ZHEJIANG UNIV
View PDF 2 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
ZHEJIANG UNIV
Filing Date
2022-12-02
Publication Date
2026-07-03

AI Technical Summary

Technical Problem

Existing PET image reconstruction methods struggle to balance speed and quality. Analytical reconstruction methods based on inverse Radon transform are fast but produce poor image quality, while iterative framework-based methods are slow and have poor generalization. Analytical algorithms combined with deep learning are fast but lack sufficient detail in the reconstructed images.

Method used

An attention SE module is added to the denoising part of FBP-Net, and the loss function during the training phase is improved. Image quality is enhanced by filtering backprojection layers and attention enhancement layers. The L1 loss function and VGG loss function are combined to enhance image details.

Benefits of technology

It improves the quality and detail of PET image reconstruction while maintaining high generalization and speed, providing a more reliable diagnostic basis.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN116309898B_ABST
    Figure CN116309898B_ABST
Patent Text Reader

Abstract

This invention discloses a dynamic reconstruction method to enhance the details of PET images. This method improves upon the existing FBP-Net by adding an SE attention module to the denoising neural network part of FBP-Net and refining the loss function during the training phase. The L2 loss function, which addresses blurred details, is replaced with an L1 loss function and a VGG loss function, allowing the network structure to focus more on reconstructing the details of the reconstructed image. This invention solves the problems of blurry and insufficient detail in FBP-Net reconstructed images, inheriting the advantages of FBP-Net's strong generalization ability while significantly improving the quality of PET reconstructed images.
Need to check novelty before this filing date? Find Prior Art