PET Image Reconstruction With Swin-Transformer Noise Regularization

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing PET image reconstruction methods face challenges in achieving high-quality images from low dose projection data, often resulting in high noise levels and radiation exposure for patients, while current deep learning approaches like CNNs struggle with long-distance dependencies and feature resolution.

Innovation Solution

A PET image reconstruction method utilizing Swin-Transformer regularization, comprising iterative modules with EM layers, Swin-Transformer layers, and pixel-to-pixel fusion, to effectively reduce noise and retain structural information, even with reduced dose data.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If the injection dose is increased to enhance PET image quality, then image quality is improved, but radiation exposure to patients increases

Engineering Contradiction:

ImprovePET image qualityVSAvoidradiation exposure

Core Design Contradiction:

Measurement precisionVSObject-affected harmful factors

Solution Approach 1:

The patent changes the parameter of injection dose from high to low, and uses algorithmic parameters (regularization strength, iteration count) to compensate for the reduced physical dose, thereby maintaining image quality while reducing radiation exposure

Inventive Principle:

Principle #35Parameter changes

Solution Approach 2:

The patent replaces the physical mechanism of increasing radiation dose with an algorithmic mechanism (Swin-Transformer regularization), substituting computational processing for physical intervention to achieve image quality improvement without additional radiation

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

2Loss of information

If CNNs are used with more layers to handle long-distance dependencies, then dependency modeling is improved, but feature resolution and fine details are lost

Engineering Contradiction:

Improvelong-distance dependency modelingVSAvoidfeature resolution

Core Design Contradiction:

Loss of informationVSManufacturing precision

Solution Approach 1:

The patent segments the image processing into two distinct components: CNN for local feature extraction and Swin-Transformer for global dependency modeling, allowing each to operate at its optimal scale without compromising the other

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The patent introduces Swin-Transformer as an intermediary component that bridges the gap between local CNN features and global image context, enabling long-distance dependency modeling while preserving fine details through its hierarchical window-based attention mechanism

Inventive Principle:

Principle #24Intermediary (Mediator)

Data Source

PatentUS12511799B2Pet image reconstruction method based on swin-transformer regularization

Publication Date: 2025.12.30 ZHEJIANG UNIV

AI summary

The present invention discloses a PET image reconstruction method based on Swin-Transformer regularization, the reconstruction model adopted is composed of several iterative modules, each iterative module consists of three parts: a EM iterative layer, a Swin-transformer based regularization layer, a pixel to pixel image fusion layer, the regularization layer is used to learn the prior information representing the image, comprising a convolution kernel used to extract the shallow features of the image, a Swin-Transformer layer used to extract the deep features of the image, and a convolution layer and a residual connection are used to fuse deep and shallow features. The image fusion layer fuses the results of EM iteration and regularization. The invention can reconstruct high quality PET images from Sinogram projection data, and greatly reduces the noise level of PET images while retaining the structural information.