Radiography Image Analysis Using Neural Network Geometric Localization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Despite the skills of radiologists, cases of misdiagnosis in radiography lead to medical errors due to the lack of a reliable method for analyzing medical images, which can result in detrimental consequences for patients.
Innovation Solution
A computer-implemented method for analyzing radiography images that involves detecting and locating remarkable geometric elements using neural networks, measuring angles between bones, and classifying anomalies, with multiple techniques for image processing and aggregation to enhance reliability.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If multiple neural network techniques are implemented for locating geometric elements, then measurement precision is improved, but device complexity increases
Solution Approach 1:
The patent segments the geometric element localization task into two distinct neural network techniques: one that determines notable points and connects them to form segments, and another that determines notable lines and finds their intersections as notable points. This segmentation allows each technique to specialize in different aspects of the problem, improving overall precision while organizing the complexity into manageable, modular components.
Solution Approach 2:
The patent merges the results from two different neural network techniques through an aggregation step that combines their outputs. By implementing both techniques and aggregating their results, the system leverages the strengths of each approach to achieve higher measurement precision than either technique could achieve alone, while the aggregation mechanism manages the complexity of integrating multiple methods.
2Productivity
If automated image analysis is implemented, then productivity is improved, but reliability may worsen due to potential algorithmic errors
Solution Approach 1:
The patent implements feedback mechanisms where the system detects the imaged area and imaging angle, uses this information to adaptively locate geometric elements, and then uses the located elements to perform measurements. The system also provides feedback by generating reports with measured values and can display these results, allowing for verification and correction if needed. This closed-loop approach maintains reliability while achieving automated productivity.
Solution Approach 2:
The patent performs preliminary actions by first detecting the imaged area and imaging angle before proceeding with geometric element localization. It also pre-processes the image by identifying notable lines and points before performing the final measurements. These preliminary steps prepare the data in advance, ensuring that the subsequent automated analysis is based on accurate, pre-validated information, thereby maintaining reliability while enabling efficient automated processing.
3Measurement precision
If detailed geometric element localization is performed, then measurement precision is improved, but loss of time increases due to processing complexity
Solution Approach 1:
The patent performs preliminary detection of the imaged area and imaging angle before the detailed geometric element localization. This preliminary action prepares the context information in advance, allowing the subsequent localization process to proceed more efficiently with pre-known constraints and parameters, reducing the overall processing time while maintaining detailed precision in the final measurements.
Solution Approach 2:
The patent segments the time-consuming localization process into two parallel neural network techniques that can be implemented independently and then aggregated. By dividing the complex localization task into two separate but complementary approaches (points-to-segments and lines-to-intersections), the system can process information through multiple pathways simultaneously, reducing the total processing time while achieving detailed precision through the aggregation of both techniques' results.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
The invention relates to a method for analyzing images taken by radiography of a part of a subject comprising at least one area to be studied with bones, comprising a step of: - detecting an area to be studied as well as the incidence with which the images are taken by radiography, - locating notable geometric elements of the area to be studied by implementing a first technique involving the determination of the position of notable points using a neural network and then the determination of notable segments as connecting the determined points or a second technique involving the determination of notable lines using a neural network and then the determination of notable points as intersection of determined notable lines, and - measuring at least one angle between two bones of the area to be studied from the location of notable geometric elements.