Medical Image Denoising via Synthetic Noise Simulation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The limited availability of medical images for training deep learning models for denoising medical images poses a challenge, as these models require a substantial amount of data to effectively reduce noise.
Innovation Solution
A noise simulator generates a larger number of medical image sets for training by simulating noise components and low-quality images, which are then used to train a deep learning model for noise extraction, enabling effective denoising of medical images.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If a deep learning model is trained using a small number of actual medical images, then the training process can be completed with limited data availability, but the noise reduction performance is insufficient due to inadequate training data quantity
Solution Approach 1:
The patent creates synthetic training data by copying and transforming existing medical images through simulated noise addition. The noise simulator generates multiple copies of original images with different noise characteristics, effectively multiplying the training dataset size without requiring additional real medical images.
Solution Approach 2:
The noise simulator is pre-trained on a small set of actual medical images to learn noise characteristics. This preliminary training enables the simulator to generate realistic noisy images that can serve as training data for the main denoising model, preparing the training dataset in advance before the actual denoising model training begins.
2Reliability
If more medical images are collected for training, then the noise reduction performance can be improved, but the data acquisition time and cost increase
Solution Approach 1:
Instead of acquiring additional real medical images, the system creates synthetic copies by adding simulated noise to existing images. This copying approach generates abundant training data instantaneously without the time-consuming process of collecting, annotating, and preparing real medical images.
Solution Approach 2:
The patent extracts noise characteristics from a small set of actual medical images using the noise simulator, then separates these learned noise patterns from the original images. This extraction process allows the generation of diverse training samples without needing to collect additional real images, saving data acquisition time.
3Measurement precision
If the noise simulator is trained to generate realistic noise, then the quality of synthetic training data improves, but the training complexity of the noise simulator increases
Solution Approach 1:
The noise simulator is trained in advance on a small dataset to learn noise characteristics before being used to generate training data for the denoising model. This preliminary training phase, while complex, is performed once and enables the simulator to efficiently generate high-quality synthetic noise for subsequent training iterations.
Solution Approach 2:
The noise simulator acts as an intermediary between the limited actual medical images and the requirement for large-scale training data. It mediates by transforming a small set of real images into a large set of synthetic noisy images, bridging the gap between data scarcity and model training requirements through learned noise patterns.
Data Source
AI summary
Disclosed is an apparatus for denoising a medical image, including: an image processing module configured to extract a noise component from a medical image for processing by inputting the medical image for the processing to a noise extraction deep learning model trained in advance, and generate a noise-removed image by subtracting the noise component from the medical image for the processing, wherein the noise extraction deep learning model is trained using a simulation noise component image generated by a noise simulator and a simulation low-quality image generated based on the simulation noise component image as a pair.