Neural Network Deconvolution for Blurred PET Histo-Images

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Conventional Positron Emission Tomography (PET) imaging produces blurred histo-images due to uncertainties in event location determinations and large compression ratios between Time-of-Flight (TOF) data and image reconstruction, leading to inferior image quality, which is exacerbated by the computational and memory-intensive nature of existing deblurring methods.

Innovation Solution

A neural network is trained using PET images reconstructed from raw data to deconvolve blurred histo-images, generating a simulated reconstructed PET image that is similar in quality and resolution to ground-truth images, while reducing processing resources and time, optionally incorporating mu-maps for improved attenuation corrections.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Loss of time

If conventional deblurring kernels are applied to single histo-image, then processing time is reduced, but image quality deteriorates

Engineering Contradiction:

Improveprocessing timeVSAvoidimage quality

Core Design Contradiction:

Loss of timeVSManufacturing precision

Solution Approach 1:

The patent merges multiple deblurred histo-images into a single reconstructed image by performing MLA back-projections of different view data to generate multiple histo-images, deblurring each individually, and then combining them. This combination approach maintains image quality while distributing the computational workload.

Inventive Principle:

Principle #5Merging (Combining)

Solution Approach 2:

The reconstruction process is segmented into multiple independent steps: generating multiple histo-images from different views, deblurring each histo-image separately using deblurring kernels, and then combining the results. This segmentation allows parallel processing and optimizes both quality and efficiency.

Inventive Principle:

Principle #1Segmentation

2Manufacturing precision

If multiple histo-images are generated and deblurred individually, then image quality is improved, but computational resources and memory requirements increase

Engineering Contradiction:

Improveimage qualityVSAvoidcomputational resources

Core Design Contradiction:

Manufacturing precisionVSDevice complexity

Solution Approach 1:

The patent applies deblurring kernels to multiple histo-images rather than a single image, using partial actions on each view's histo-image. This approach improves overall image quality by leveraging information from multiple angles while managing computational resources through selective processing of each view.

Inventive Principle:

Principle #16Partial or excessive action

3Productivity

If TOF data compression ratio is increased, then data processing efficiency is improved, but image resolution deteriorates

Engineering Contradiction:

Improvedata processing efficiencyVSAvoidimage resolution

Core Design Contradiction:

ProductivityVSManufacturing precision

Solution Approach 1:

The patent introduces multiple histo-images generated from different view data as intermediaries between the compressed TOF data and the final reconstructed image. These intermediate representations preserve more information than a single compressed reconstruction, allowing better resolution recovery while maintaining processing efficiency.

Inventive Principle:

Principle #24Intermediary (Mediator)

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

The neural network approach produces superior quality PET images faster and with fewer resources than conventional methods, effectively addressing the blurring issues and improving image resolution without the computational intensity of existing systems.

Implementation Method 1

A neural network is trained using PET images reconstructed from raw data to deconvolve blurred histo-images, generating a simulated reconstructed PET image

Methodology Applied

Scientific EffectNeural network deconvolution:

Implementation Method 2

Time-of-flight (TOF) PET measures the difference between the detection times of the two gamma photons arising from the annihilation event. This difference may be used to estimate a particular position along the LOR at which the annihilation event occurred

Methodology Applied

Scientific EffectTime-of-flight: Time of Flight

Implementation Method 3

Radioactive decay of the radionuclide generates positrons, which eventually encounter electrons and are annihilated thereby. Annihilation produces two gamma photons which travel in approximately opposite directions

Methodology Applied

Scientific EffectRadioactive decay: Radioactive Decay

Data Source

PatentUS11164344B2PET image reconstruction using TOF data and neural network

Publication Date: 2021.11.02 SIEMENS MEDICAL SOLUTIONS USA INC

AI summary

A system and method include execution of a first scan to acquire a first PET dataset, back-projection of the first PET dataset to generate a first histo-image, input of the first histo-image to a trained neural network, and reception of a first output image from the trained neural network.