Medical Image ROI Inference for Stable Abnormal Region Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing techniques for automatic or semi-automatic detection of abnormal regions in medical images face challenges in maintaining a stable detection rate, particularly during the second inference process, and do not maximize the detection rate effectively.
Innovation Solution
An image processing apparatus that performs a two-step inference process, using a machine learning model for initial region extraction followed by a second-step identification process, with the setting of multiple processing target regions and optional termination conditions to enhance accuracy and reduce computational cost.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If a two-step inference process is used to increase detection rate, then the detection rate improves, but the computational cost and processing time increase
Solution Approach 1:
The patent applies preliminary action by performing a first inference process to extract candidate abnormal regions before conducting the second inference process. This preliminary extraction step prepares the data in advance, allowing the second step to focus only on relevant regions rather than processing the entire image, thus reducing overall processing time while maintaining high detection rate.
Solution Approach 2:
The patent segments the image processing into two distinct inference steps: first inference for candidate region extraction and second inference for final abnormal region determination. This segmentation allows each step to be optimized independently, with the first step filtering out obvious non-abnormal areas and the second step focusing computational resources on promising candidates, thereby reducing total processing time.
2Reliability
If the processing target region is expanded to increase detection rate, then the detection rate improves, but the computational overhead increases
Solution Approach 1:
The patent applies local quality by setting different processing strategies for different regions of the image. The first inference process identifies candidate regions with different characteristics, and the second inference process applies targeted processing only to these candidate regions. This localized approach ensures high detection rate within candidate regions while avoiding unnecessary computational overhead in non-candidate areas.
Solution Approach 2:
The patent uses partial action by performing the computationally intensive second inference process only on candidate regions identified by the first inference, rather than on the entire image. This partial application of the detailed analysis step reduces computational overhead significantly while maintaining high detection rate, as the second step is applied only where needed based on first-step results.
3Reliability
If multiple processing target regions are set to maximize detection rate, then the detection rate improves, but the complexity of region management increases
Solution Approach 1:
The patent implements feedback by using the results of the first inference process to inform and guide the second inference process. The candidate regions identified in the first step serve as feedback that directs the focus and scope of the second step. This feedback mechanism simplifies region management by automatically determining which regions require detailed analysis based on first-step outcomes, reducing the complexity of manual region configuration.
Solution Approach 2:
The patent applies dynamics by making the processing target regions adaptive rather than fixed. The candidate regions for the second inference are dynamically determined based on the results of the first inference process. This dynamic region selection automatically adjusts the processing scope to match the actual content of each image, simplifying region management while maximizing detection rate across varying image types and abnormalities.
Data Source
AI summary
An image processing apparatus includes one or more memories that stores instructions and one or more processors that, upon execution of the instructions, operates as: an image acquisition unit, a region-of-interest setting unit, a processing target region setting unit, and an identification unit. The image acquisition unit acquires an image. The region-of-interest setting unit set a region of interest in the image. The processing target region setting unit sets a plurality of processing target regions for the region of interest. The identification unit performs an identification process on the plurality of processing target regions. The identification unit terminates the identification process when a predetermined termination condition is met.