Side Lung Field Segmentation With User Correction for Area Accuracy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image processing techniques for lung field region recognition in X-ray images, such as those described in JP 2019-122449A and JP 2020-171427A, fail to accurately distinguish left and right lung field regions, leading to low area accuracy in lung volume calculations.
Innovation Solution
An image processing apparatus and method that identifies and calculates the area of first and second side lung field regions using deep learning, allows user interaction for region change, and includes a changer to enhance accuracy through user operation, utilizing a hardware processor and a changer to adjust lung field regions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of time
If automatic lung field region identification is performed using existing image processing techniques, then the operation time is reduced, but the area accuracy of lung field regions deteriorates
Solution Approach 1:
The patent segments the lung field region identification process into two distinct phases: automatic identification using deep learning to quickly locate the lung field region, and manual correction allowing users to adjust the identified region. This segmentation enables the system to benefit from both automated speed and manual precision, resolving the contradiction between operation time and area accuracy.
Solution Approach 2:
The patent implements a feedback mechanism where the automatically identified lung field region is displayed to the user for verification and correction. The user's manual adjustments serve as feedback to refine the automatic identification results, creating a closed-loop system that improves accuracy while maintaining efficiency.
2Measurement precision
If manual correction of lung field regions is allowed, then the area accuracy is improved, but the operation time increases
Solution Approach 1:
The patent performs preliminary automatic identification of the lung field region using deep learning before allowing manual correction. This preliminary action provides a high-quality initial result that is likely close to the final accurate region, minimizing the amount of manual adjustment needed and thereby reducing the time penalty associated with manual correction.
3Productivity
If deep learning is used for automatic lung field region identification, then the productivity is improved, but the measurement precision deteriorates
Solution Approach 1:
The patent merges two different approaches - deep learning-based automatic identification and manual correction - into a unified system. The deep learning component handles the majority of cases efficiently, while the manual correction component handles edge cases or cases requiring higher precision, creating a hybrid system that achieves both high productivity and high measurement precision.
Data Source
AI summary
An image processing apparatus, including: a hardware processor that obtains a side image of a lung field radiographed from a side of a subject, identifies a first side lung field region and a second side lung field region, from the side image, causes a display to display at least one of the identified first side lung field region and second side lung field region, and calculates at least one of an area of the first side lung field region and an area of the second side lung field region, based on the first side lung field region and/or the second side lung field region; and a changer that is capable of changing at least one side lung field region between the first side lung field region and the second side lung field region through operation by a user.