Neural Network Estimation of Emphasized Compositions from 2D Radiography
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for estimating images that emphasize specific compositions, such as bone parts, in radiography struggle with accuracy, particularly when using simple two-dimensional images.
Innovation Solution
A neural network-based estimation device that learns from composite two-dimensional images derived from three-dimensional CT images and emphasis images, using attenuation coefficients and projection techniques to enhance the accuracy of composition emphasis in radiography.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If simple two-dimensional radiation images are used for estimation, then the imaging process is simple and fast, but the accuracy of emphasizing specific compositions is insufficient
Solution Approach 1:
The patent applies preliminary action by pre-processing the simple two-dimensional radiation image to generate an initial estimation image before applying the neural network. This preliminary processing step prepares the data in a format that enhances the neural network's ability to accurately estimate specific compositions, thereby improving measurement precision while managing processing complexity through structured pre-computation.
Solution Approach 2:
The patent introduces an intermediary approach by using a neural network as a mediator between the simple two-dimensional radiation image and the final emphasized composition image. The neural network learns the complex mapping relationship from training data, acting as an intelligent intermediary that transforms simple images into accurate composition-emphasized images without requiring direct complex processing of the original radiation image.
2Adaptability or versatility
If traditional energy subtraction processing is used, then the processing method is straightforward, but the derivation of specific compositions in soft parts is limited
Solution Approach 1:
The patent applies parameter changes by training the neural network with multiple types of teacher data representing different compositions (bone parts, soft parts, fat, muscle). The network learns to adjust its processing parameters based on the input simple image characteristics, enabling it to accurately derive different composition types with high precision, thereby improving both adaptability and measurement precision simultaneously.
Solution Approach 2:
The patent uses a composite approach by combining multiple types of teacher data (different composition images) during the neural network training process. This composite training data strategy enables the network to learn comprehensive composition derivation capabilities, improving versatility to handle various tissue types while maintaining high accuracy through diverse training examples.
3Measurement precision
If learned models are trained with basic radiation images, then the training process is simple, but the estimation accuracy of emphasized compositions is insufficient
Solution Approach 1:
The patent applies the copying principle by creating synthesized training data through generating simple two-dimensional radiation images and their corresponding emphasized composition images from three-dimensional CT data. This copying approach allows the creation of large quantities of realistic training pairs without requiring actual clinical data, thereby improving estimation accuracy while managing the quantity of training data through efficient synthetic data generation.
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
The solution enables high-accuracy estimation of emphasized compositions like bone parts, soft parts, muscle, and fat in radiography, improving image processing outcomes.
Implementation Method 1
which is derived by combining a three-dimensional CT image of the subject, and an emphasis image for learning in which the specific composition of the subject is emphasized
Implementation Method 2
the composite two-dimensional image may be derived by deriving an attenuation coefficient of radiation for a composition at each position on a three-dimensional space, and projecting the CT image in a predetermined direction
Data Source
AI summary
An estimation device includes at least one processor, in which the processor functions as a learned neural network that derives a result of estimation of at least one emphasis image in which a specific composition of a subject including a plurality of compositions is emphasized from a simple two-dimensional image acquired by simply imaging the subject. The learned neural network is learned by using, as teacher data, a composite two-dimensional image representing the subject, which is derived by combining a three-dimensional CT image of the subject, and an emphasis image for learning in which the specific composition of the subject is emphasized, which is derived from the CT image.