Modular Neural Network for Low-Dose CT Image Denoising
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Low-dose CT images often suffer from noise and artifacts due to reduced radiation exposure, which can impair diagnostic performance and treatment outcomes, and existing denoising models struggle to generalize across different radiation dose levels.
Innovation Solution
A modularized adaptive processing neural network (MAP-NN) system comprising multiple trained neural network modules cascaded in series, allowing for incremental denoising and selection of optimal intermediate output images by a domain expert, with the ability to determine an optimum mapping depth and weighted sum of images to enhance image quality.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Object-affected harmful factors
If radiation dosage is reduced for CT scans, then patient exposure to x-ray radiation is decreased, but image quality deteriorates with increased noise and artifacts
Solution Approach 1:
The patent introduces a denoising model as an intermediary between the low-dose CT image acquisition and the final diagnostic image. This model processes the noisy low-dose images to remove artifacts and noise, effectively mediating between the reduced radiation dose and the requirement for high image quality. The denoising model acts as a bridge that allows low-dose imaging while maintaining diagnostic image quality through computational processing.
2Adaptability or versatility
If a single denoising model is used, then the system is simple, but the model cannot generalize across different radiation dose levels
Solution Approach 1:
The patent segments the denoising system into multiple specialized denoising models, each trained for specific radiation dose levels or image quality requirements. Instead of using one general model, the system divides the denoising task across multiple models that can be selectively applied. This segmentation allows each model to be optimized for its specific domain while the overall system gains versatility across different dose levels.
Solution Approach 2:
The patent implements a dynamic model selection mechanism that adapts which denoising model to use based on the input image characteristics and desired output. The system dynamically chooses the appropriate denoising model rather than using a fixed approach, allowing adaptation to different radiation dose levels and image quality requirements. This dynamic selection enables the system to handle varying conditions without requiring a completely different model for each scenario.
3Adaptability or versatility
If multiple denoising models are trained for different dose levels, then adaptability to various radiation levels improves, but training time and computational resources increase
Solution Approach 1:
The patent applies preliminary action by pre-training multiple denoising models on different radiation dose levels before actual clinical use. These models are prepared in advance and stored for rapid deployment. When a low-dose CT image needs processing, the pre-trained models are already available for immediate application without requiring training at the moment of need. This preliminary preparation reduces the time loss during actual diagnostic workflows.
Solution Approach 2:
The patent utilizes parameter changes by training denoising models with different hyperparameters optimized for specific radiation dose levels. Each model is configured with parameters tailored to its target dose range, allowing efficient specialization without requiring complete retraining when switching between dose levels. This parameter-based adaptation enables the system to handle multiple dose levels efficiently while reducing overall training time.
Data Source
AI summary
A system for enhancing a low-dose (LD) computed tomography (CT) image is described. The system includes a modularized adaptive processing neural network (MAP-NN) apparatus and a MAP module. The MAP-NN apparatus is configured to receive a LDCT image as input. The MAP-NN apparatus includes a number, T, trained neural network (NN) modules coupled in series. Each trained NN module is configured to generate a respective test intermediate output image based, at least in part, on a respective received test input image. Each test intermediate output image corresponds to an incrementally denoised respective received test input image. The MAP module is configured to identify an optimum mapping depth, D, based, at least in part, on a selected test intermediate output image, the selected test intermediate output image selected by a domain expert. The mapping depth, D, is less than or equal to the number, T.