PET Image Reconstruction Using Deep Learning Attenuation Correction

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Current positron emission tomography (PET) systems require additional scans like CT or MRI to obtain high-quality attenuation images, leading to increased radiation exposure and difficulties in image reconstruction due to crosstalk artifacts and noise.

Innovation Solution

A deep learning algorithm-based positron emission tomography system that generates high-quality attenuation and emission images simultaneously without additional scanning, using a collection unit for PET sinograms, a generation unit for MLAA with TOF, and a learning unit to adjust weights and generate final images through a convolutional neural network.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If additional CT or MRI scan is performed to obtain attenuation image, then image quality is improved, but radiation exposure to patient increases

Engineering Contradiction:

Improveimage qualityVSAvoidradiation exposure

Core Design Contradiction:

Measurement precisionVSObject-affected harmful factors

Solution Approach 1:

The patent combines the attenuation image acquisition and emission image acquisition into a single PET scan process using the MLAA algorithm. Instead of performing separate CT/MRI scans for attenuation correction, the system simultaneously reconstructs both attenuation and emission images from the PET data, eliminating the need for additional scanning and reducing radiation exposure while maintaining diagnostic image quality

Inventive Principle:

Principle #5Merging (Combining)

Solution Approach 2:

The PET scanner is made multi-functional by enabling it to perform both attenuation imaging and emission imaging through the MLAA reconstruction algorithm. The single PET acquisition process serves dual purposes: obtaining the emission data for functional imaging and deriving the attenuation data for correction, replacing the need for separate dedicated attenuation scanning devices

Inventive Principle:

Principle #6Universality (Multi-functionality)

2Loss of time

If MLAA with TOF algorithm is used to simultaneously obtain attenuation and emission images, then additional scanning is eliminated, but crosstalk artifacts and noise increase

Engineering Contradiction:

Improvescanning timeVSAvoidimage quality

Core Design Contradiction:

Loss of timeVSReliability

Solution Approach 1:

The patent introduces a deep learning-based post-processing step as an intermediary between the MLAA reconstruction and final image output. This intermediate processing stage specifically targets and removes crosstalk artifacts and noise while preserving the time-efficient simultaneous acquisition benefit, acting as a cleanup mechanism that enhances image reliability without requiring additional scanning time

Inventive Principle:

Principle #24Intermediary (Mediator)

Solution Approach 2:

The patent replaces traditional artifact reduction methods with deep learning-based image processing. Instead of using conventional filtering or correction algorithms, a neural network is trained to recognize and eliminate crosstalk artifacts and noise patterns, substituting mechanical/mathematical processing with intelligent algorithms that achieve better artifact suppression while maintaining the efficiency of simultaneous reconstruction

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

Applied Scientific Principles

This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.

Function Achieved in This Case

This approach minimizes radiation exposure, reduces reconstruction time and cost, and improves image quality by effectively addressing noise and crosstalk artifacts, achieving high similarity with CT-derived images.

Implementation Method 1

positron emission tomography (PET) is a nuclear medicine testing method that can show physiological/chemical and functional images of a body in three dimensions using radiopharmaceuticals emitting positron

Methodology Applied

Scientific EffectPositron emission: Radioactive Decay

Implementation Method 2

detecting two disappearing radiations (gamma rays) emitted from the positron

Methodology Applied

Scientific EffectGamma ray emission: Radiation

Implementation Method 3

simultaneous reconstruction algorithm (Maximum Likelihood reconstruction of Attenuation and Activity with Time-Of-Flight, MLAA with TOF)

Methodology Applied

Scientific EffectTime-of-flight: Time of Flight

Implementation Method 4

applying the learned deep learning algorithm to the input image to generate and provide a final attenuation image

Methodology Applied

Scientific EffectDeep learning: Image Processing

Data Source

PatentUS11756241B2Positron emission tomography system and image reconstruction method using the same

Publication Date: 2023.09.12 SEOUL NATIONAL UNIVERSITY R&DB FOUNDATION

AI summary

Disclosed are a positron emission tomography system and an image reconstructing method using the same and the positron emission tomography system includes: a collection unit collecting a positron emission tomography sinogram (PET sinogram); an image generation unit applying the positron emission tomography sinogram to an MLAA with TOF and generating a first emission image and a first attenuation image and a NAC image reconstructed without attenuation correction; and a control unit selecting at least one of the first emission image, the first attenuation image and the NAC image generated by the image generation unit as an input image and generating and providing a final attenuation image by applying the input image to the learned deep learning algorithm.