Medical Image Superresolution by Sensor Mode-Matched CNN Models
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing superresolution processing methods using convolutional neural networks (CNNs) for medical images face a decrease in quality when using low-resolution images generated by different sensor operation methods, leading to loss of detailed information in reduced-resolution radiographic images.
Innovation Solution
A medical-image processing apparatus that selects a learning model based on the operation mode of the sensor used for image capture, generating a medical image of higher resolution through supervised learning with CNNs, matching the generation method of the input image to that used in training, thereby improving resolution and reducing noise.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If superresolution processing using a single CNN learning model is used for all sensor operation modes, then the processing method is simple, but the quality of superresolution images decreases when the sensor operation mode does not match the training data
Solution Approach 1:
The patent segments the single learning model into multiple learning models, each trained for a specific sensor operation mode (e.g., binning modes 1×1, 2×2, 3×3). The system selects and applies the appropriate learning model based on the actual sensor operation mode, ensuring high-quality superresolution processing for each mode while maintaining manageable system complexity through modular organization.
2Manufacturing precision
If multiple learning models are used for different sensor operation modes, then the superresolution image quality is maintained, but the system complexity increases
Solution Approach 1:
The patent introduces an operation mode detection unit as an intermediary that automatically identifies the sensor's current operation mode and selects the corresponding learning model. This mediator component manages the complexity of having multiple learning models by providing an automated selection mechanism, eliminating the need for manual configuration and ensuring the correct model is applied without increasing operational complexity for the user.
3Object-affected harmful factors
If reduced-resolution images are used to reduce radiation exposure, then radiation dose is reduced, but detailed information such as lesion information is lost
Solution Approach 1:
The patent applies superresolution processing as a preliminary action to low-resolution medical images obtained with reduced radiation exposure. By using operation mode-matched learning models, the system reconstructs high-resolution images that recover detailed information such as lesion characteristics and anatomical structures, enabling accurate diagnosis while maintaining the radiation reduction benefit.
Data Source
AI summary
A medical-image processing apparatus according to the present invention includes an obtaining unit configured to obtain a medical image obtained by capturing an image of an examinee and a generation unit configured to input the medical image to a learning model selected based on an operation mode of a sensor at the image capturing to generate a medical image of a higher resolution than a resolution of the medical image.