Neural Network for Radiation Image Composition Estimation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current radiation imaging techniques face challenges in accurately estimating body thickness and composition, particularly when objects like grids or imaging table components interfere with the radiation, causing beam hardening and scattering that reduce image contrast and accuracy.
Innovation Solution
A learning device and method that utilize a neural network trained with data from standard objects to estimate body thickness and composition by accounting for the energy characteristics and materials of objects between the subject and radiation detector, allowing for precise derivation of body thickness, bone mineral density, and muscle mass through energy subtraction processing.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Object-affected harmful factors
If scattered ray removal processing is performed using radiation attenuation coefficient, then scattered ray component can be removed, but accuracy is reduced due to beam hardening from objects like grids and imaging tables
Solution Approach 1:
A neural network is introduced as an intermediary between the radiation image and the body thickness/composition estimation. The neural network is trained to recognize and compensate for beam hardening effects caused by objects like grids and imaging tables, thereby removing the harmful influence of these objects on measurement accuracy without requiring their precise characterization.
Solution Approach 2:
The system transforms the radiation image data through the neural network, which learns optimal parameter transformations during training. The neural network adjusts internal parameters (weights and biases) to map input radiation images to accurate body thickness and composition values, effectively changing the parameter representation to account for beam hardening effects.
2Measurement precision
If beam hardening from objects between subject and detector is considered, then accuracy of composition estimation improves, but complexity of the system increases
Solution Approach 1:
The neural network performs self-learning and self-adjustment during the training phase. By training on diverse radiation images including those with various objects like grids and imaging tables, the neural network automatically adapts to compensate for beam hardening effects without requiring explicit programming or complex calibration procedures for each object type.
3Measurement precision
If neural network is trained with standard objects to account for object characteristics, then estimation accuracy improves, but training data requirements and processing time increase
Solution Approach 1:
The neural network is trained in advance using a comprehensive dataset of standard objects and radiation images. This preliminary training action prepares the neural network to handle various beam hardening scenarios, so that during actual medical imaging, the pre-trained network can quickly and accurately estimate body thickness and composition without requiring additional real-time processing or object-specific calibration.
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
Enables accurate estimation of body thickness and composition by constructing a neural network that accounts for the energy characteristics and materials of objects in the imaging setup, improving the accuracy of radiation image processing and composition analysis.
Implementation Method 1
constructs, by subjecting a neural network to machine learning using teacher data consisting of learning data and correct answer data, a trained neural network that outputs at least one of a composition of a target subject or a body thickness of the target subject
Implementation Method 2
a radiation image acquiring unit that acquires at least one target radiation image of a target subject
Implementation Method 3
energy subtraction processing using two radiation images obtained by irradiating a subject with two types of the radiation having different energy distributions
Implementation Method 4
scattered ray removal processing for removing a scattered ray component included in the radiation image is performed
Data Source
AI summary
By subjecting a neural network to machine learning using teacher data consisting of learning data and correct answer data, a trained neural network that outputs at least one of a composition of a target subject or a body thickness of the target subject in a case in which at least one target radiation image of the target subject is input is constructed. The learning data includes a standard image acquired by irradiating a standard object of which a thickness and a material are known with radiation in a state in which an object is interposed between the standard object and a radiation detector, an energy characteristic of the radiation, a thickness and a material of the object, and an imaging condition in a case in which the standard image is acquired. The correct answer data includes at least one of a composition or a body thickness of a subject derived from at least one radiation image of the subject by using the standard image, the energy characteristic of the radiation, the thickness and the material of the object, and the imaging condition.