Apriori Guidance Network for Medical Image Synthesis

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Current medical image synthesis technologies face challenges in generating high-quality CT and MRI images from PET images due to limitations in data availability, error accumulation in serial multitask learning, and specificity issues in integrated multitask learning, leading to suboptimal image conversion and increased workload.

Innovation Solution

An apriori guidance network is introduced, utilizing a generator and discriminator with an apriori guidance module to convert PET images into CT or MRI images using a generative adversarial mechanism, leveraging apriori features to guide the network and improve image synthesis quality, and employing alternate learning strategies to reduce error and increase efficiency.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Ease of manufacture

If serial multitask learning is used to convert PET images to CT and then to MRI images, then the conversion process can be broken down into manageable steps, but error accumulation occurs and image quality deteriorates

Engineering Contradiction:

Improveconvertibility of image synthesis taskVSAvoidimage quality

Core Design Contradiction:

Ease of manufactureVSManufacturing precision

Solution Approach 1:

The patent segments the multitask learning problem into independent parallel tasks (PET to CT, PET to MRI) rather than serial composition. Each task is handled by a separate network module that directly processes the original PET image, preventing error accumulation from sequential transformations while maintaining task manageability through modular architecture.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

Instead of the conventional serial approach (PET→CT→MRI), the patent inverts the logic by implementing parallel direct transformations from the source modality (PET) to each target modality (CT, MRI) simultaneously. This inversion eliminates the intermediate error propagation step while achieving the same ultimate goal of multi-modal image generation.

Inventive Principle:

Principle #13The other way round (Inversion)

2Productivity

If integrated multitask learning is used to generate both CT and MRI images from PET images simultaneously, then efficiency is improved, but specificity issues arise and image synthesis quality decreases

Engineering Contradiction:

Improveefficiency of image synthesisVSAvoidimage synthesis quality

Core Design Contradiction:

ProductivityVSManufacturing precision

Solution Approach 1:

The patent applies local quality by giving each task-specific module its own dedicated network structure optimized for its target modality (CT or MRI). Each module processes the input with modality-specific features and parameters, ensuring high synthesis quality for each task while maintaining overall system efficiency through parallel execution. This allows each submodule to specialize rather than sharing generic processing.

Inventive Principle:

Principle #3Local quality

Solution Approach 2:

The system dynamically adapts by selectively activating only the required task module based on the desired output modality. When CT images are needed, only the PET-to-CT module is activated; when MRI images are needed, only the PET-to-MRI module is activated. This dynamic task selection maintains high quality by avoiding interference from unrelated tasks while preserving efficiency through on-demand processing.

Inventive Principle:

Principle #15Dynamics

3Productivity

If joint multitask learning is used to implement generation from PET images to both CT and MRI images, then both images can be generated simultaneously, but data availability becomes a limiting factor due to lack of matched PET-CT-MR data pairs

Engineering Contradiction:

Improvesimultaneous generation of multiple modalitiesVSAvoidavailability of training data

Core Design Contradiction:

ProductivityVSQuantity of substance

Solution Approach 1:

The patent creates a universal framework where a single PET image can serve as input for multiple independent transformation tasks. The system is designed to handle different target modalities (CT, MRI) with the same input modality (PET), making the overall system multi-functional. Each task module is trained on its specific data pairs (PET-CT or PET-MRI) independently, allowing the system to leverage existing data without requiring rare triple-modality datasets.

Inventive Principle:

Principle #6Universality (Multi-functionality)

Data Source

PatentUS11915401B2Apriori guidance network for multitask medical image synthesis

Publication Date: 2024.02.27 SHENZHEN INST OF ADVANCED TECH

AI summary

An apriori guidance network for multitask medical image synthesis is provided. The apriori guidance network includes a generator and a discriminator, wherein the generator includes an apriori guidance module configured to convert an input feature map into a target modal image pointing to a target domain according to an apriori feature, and the apriori feature is a deep feature of the target modal image. The generator is configured to generate a corresponding target domain image by taking the apriori feature of the target modal image and source modal image data as an input. The discriminator is configured to discriminate an authenticity of the target domain image outputted by the generator.