Multi-Model Image Reconstruction for Low-Dose Scan Quality
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Medical imaging technologies face challenges in obtaining sufficient data, leading to high noise and reduced image quality in scenarios like low-dose imaging and high-temporal resolution scans, necessitating improved image processing methods.
Innovation Solution
A system utilizing multiple trained machine learning models for image optimization, including noise reduction, contrast improvement, resolution enhancement, and artifact correction, generates intermediate images and a final target image through a series of reconstruction, optimization, and fusion units.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Object-affected harmful factors
If imaging data is reduced to enable low-dose imaging or high-temporal resolution scanning, then radiation exposure is reduced or temporal resolution is improved, but image quality deteriorates due to high noise and insufficient data
Solution Approach 1:
The patent introduces a machine learning model as an intermediary between the insufficient imaging data and the final reconstructed image. The model learns from training data the relationship between low-dose/undersampled images and high-quality reference images, enabling it to generate high-quality output images even when input data is limited, thus resolving the contradiction between reduced radiation exposure and maintained image quality
Solution Approach 2:
The system performs preliminary training of the machine learning model using paired training data (low-dose images and corresponding high-dose reference images) before actual imaging. This preliminary learning phase enables the model to acquire noise reduction and image enhancement capabilities in advance, allowing it to produce high-quality images from limited data during actual low-dose or high-temporal resolution scanning without requiring additional data collection
2Quantity of substance
If imaging data is reduced for ultra-long or ultra-short half-life drugs, then drug dosage is reduced, but image quality deteriorates due to insufficient imaging data
Solution Approach 1:
The machine learning model serves as an intermediary that compensates for the insufficient imaging data caused by reduced drug dosage. By learning the mapping between low-quality images (from ultra-long or ultra-short half-life drugs) and high-quality reference images during training, the model can generate diagnostically useful images even when the actual drug dosage is minimized, thus resolving the contradiction between reduced drug quantity and maintained image quality
3Manufacturing precision
If multiple image optimization dimensions are processed using multiple machine learning models, then image quality is comprehensively improved, but system complexity increases
Solution Approach 1:
The patent divides the image optimization task into multiple independent dimensions (noise reduction, contrast improvement, resolution enhancement, artifact correction), with each dimension handled by a separate machine learning model. This segmentation allows each model to specialize in a specific optimization aspect, improving overall image quality while maintaining manageable system complexity through modular architecture and independent model training
Solution Approach 2:
The system employs multiple machine learning models that can be applied to various imaging modalities and optimization needs. Each model is trained on specific training data to achieve universal applicability across different imaging scenarios, allowing the system to handle multiple optimization dimensions (noise, contrast, resolution, artifacts) simultaneously through a unified multi-model framework
Data Source
AI summary
The present disclosure relates to systems and methods for image processing. The systems may acquire imaging data. The systems may generate multiple intermediate images based on the imaging data by performing image optimization of multiple image optimization dimensions using a plurality of trained machine learning models. The systems may generate a target image based on the intermediate images.