Adaptive Circular ROI Texture Analysis for Ischemic Stroke Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current computer-aided detection (CAD) systems for ischemic stroke, particularly lacunar stroke, face challenges in early detection due to small lesion size and difficulty in identifying subtle changes in CT images, leading to missed diagnoses even with high-resolution diffusion-weighted imaging.
Innovation Solution
A method and system that preprocess CT images by removing bone and artifacts, generating circular adaptive regions of interest, calculating texture attributes, and comparing them between sides of the brain using a binary mask and texture analysis to enhance detection sensitivity and specificity.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If high-resolution diffusion-weighted imaging is used to detect lacunar stroke, then image resolution is improved, but detection accuracy remains insufficient due to small lesion size
Solution Approach 1:
The patent divides the brain image into multiple circular adaptive regions of interest (ROIs) with different radii. By segmenting the analysis into multiple concentric circular zones centered at suspected lesion locations, the method enables detailed texture analysis at different spatial scales, improving detection of small lacunar lesions while reducing false positives from larger artifacts
Solution Approach 2:
The patent applies different analysis strategies to different regions: circular adaptive ROIs are generated with centers at hypodense areas, and the radius of each circle is adaptively adjusted based on local image characteristics. This local adaptation allows the system to optimize detection sensitivity for each specific lesion location and size, rather than applying a uniform analysis approach throughout the entire image
2Reliability
If circular adaptive regions of interest with small radius are used to detect small lesions, then detection sensitivity for small lesions is improved, but false positive rate increases due to noise
Solution Approach 1:
The patent dynamically adjusts the radius of circular adaptive ROIs based on local image characteristics. The radius is not fixed but adaptively determined by analyzing the local texture and intensity variations. This dynamic adjustment allows the system to expand the ROI when noise is detected (reducing false positives) while maintaining small radii in clear regions (preserving sensitivity for small lesions)
Solution Approach 2:
The patent employs feedback mechanisms where the texture analysis results from initial small-radius circular ROIs are used to guide subsequent analysis. If a region shows ambiguous results or high noise characteristics, the system adjusts the radius and re-analyzes, using feedback from the initial detection to optimize the final classification and reduce false positives
3Measurement precision
If multiple texture attributes are calculated for each region, then detection accuracy is improved, but computational complexity increases
Solution Approach 1:
The patent calculates multiple texture attributes (such as gray-level co-occurrence matrix features, local binary patterns, and other texture metrics) for each circular adaptive ROI, but only for regions that pass initial screening criteria. By applying full texture analysis only to suspicious regions rather than the entire image, the system achieves high detection accuracy while limiting computational complexity through selective processing
Data Source
AI summary
A method for assisting diagnosis of stroke by image analysis, the method comprising obtaining a scanned brain image of a patient, transforming the scanned brain image into a digitized brain image, removing bone and other artifacts from the digitized brain image, generating at least one circular adaptive region of interest on one side of the brain image, generating a binary mask of the circular adaptive region of interest, calculating the percentage of zeros from the binary mask within the circular adaptive region of interest, locating at least one corresponding circular adaptive region of interest on the other side of the brain image, and comparing the circular adaptive region of interest with the corresponding circular adaptive region on interest of the other side of the brain based on a plurality of texture attributes.