Geodesic Active Contour Lesion Border Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Early detection of malignant melanoma is challenging due to variations in lesion shapes, sizes, colors, and the presence of artifacts like hair and uneven illumination, which complicates accurate segmentation in dermoscopy images.
Innovation Solution
A computer-based lesion segmentation method that utilizes geodesic active contours (GAC) and level set methods to automatically identify the border of skin lesions by transforming RGB images into grayscale and blue component planes, smoothing, binarizing, and initializing contours to avoid noise and improve accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional segmentation methods are used on dermoscopy images, then the processing is simpler, but the segmentation accuracy deteriorates due to artifacts like hair and uneven illumination
Solution Approach 1:
The patent applies preliminary actions by performing multiple smoothing operations and color plane extractions before the main segmentation process. The grayscale image is smoothed first, then the blue component plane is extracted and smoothed again, and finally the smoothed grayscale image is extracted from the smoothed blue component plane. These preliminary processing steps prepare the images to reduce the impact of artifacts like hair and uneven illumination before contour detection, thereby improving segmentation accuracy without overwhelming complexity.
Solution Approach 2:
The patent uses intermediary elements by introducing multiple processed image planes as intermediaries between the original image and the final segmentation. The smoothed grayscale image, smoothed blue component plane, and the newly extracted image serve as intermediary representations that filter out noise and artifacts. The geodesic active contour algorithm then operates on these intermediary images rather than the original noisy image, improving accuracy while managing complexity through staged processing.
2Measurement precision
If multiple image processing steps are applied to improve segmentation accuracy, then the measurement precision improves, but the processing time increases
Solution Approach 1:
The patent applies segmentation by dividing the image processing into distinct stages: grayscale conversion and smoothing, blue component extraction and smoothing, combined image extraction, and final contour detection. Each stage processes specific aspects of the image independently, allowing for optimized computation at each step rather than processing the entire image uniformly. This staged segmentation of processing tasks improves border detection accuracy while managing processing time through efficient division of labor.
3Reliability
If conventional active contour methods are used, then the algorithm is simpler, but the contour gets stuck on noise and artifacts
Solution Approach 1:
The patent applies preliminary actions by performing extensive smoothing operations on both the grayscale image and the blue component plane before the contour detection. The grayscale image is convolved with a spatial filter, and the blue component plane is also smoothed. Additionally, the blue component is extracted from the smoothed grayscale image to create a new processed plane. These preliminary smoothing and extraction operations reduce noise and artifacts before the geodesic active contour algorithm runs, preventing the contour from getting stuck on noise while maintaining algorithmic stability.
Data Source
AI summary
A method for automatic segmentation of dermoscopy skin lesion images includes automating a Geodesic Active Contour (GAC) initialization to be sufficiently large to encompass the lesion yet lie near the actual lesion contour. In addition, a new image plane is found by transforming the original RGB image to a smoothed image that allows the GAC to move without sticking on the minimum local energy. This method can eliminate the need for separate hair or noise removal algorithms. The method may include extraction of designated color planes to improve the initial contour and to automatically correct false segmentations. The method includes an algorithm to correct false protuberances and/or false inlets that may be present in the GAC border. A method is given to increase likelihood of including more actual lesion area. The method yields multiple border choices which may be presented to a classifier for optimal border selection.