Hue-texture-embedded region-based model for magnifying endoscopy with narrow-band imaging image segmentation based on...

Area of Science:

• Medical Imaging
• Computer Vision
• Gastroenterology

Background:

• Magnification endoscopy with narrow-band imaging (ME-NBI) is crucial for detecting gastrointestinal diseases, including precancerous lesions and gastric cancer.
• Traditional computer vision methods struggle with the unique visual characteristics of ME-NBI images, necessitating advanced techniques.
• Reliable computer-assisted analysis of ME-NBI images is needed to enhance disease detection accuracy.

Purpose of the Study:

• To develop and evaluate a computer-assisted methodology for analyzing ME-NBI images.
• To improve the segmentation accuracy of pathological abnormalities in the gastrointestinal tract.
• To enhance the detection of gastric cancer and its precursor lesions using ME-NBI imaging.

Main Methods:

• Developed a Hue-texture-embedded model by integrating hue and texture energy functionals into the Chan-Vese framework.
• Incorporated a global hue energy functional in the H channel (HSI color space) for color information.
• Utilized a texture energy functional in the S channel, extracting local microvascular textures via adaptive thresholding.

Main Results:

• The Hue-texture-embedded model achieved an average F-measure of 0.61 and a false positive rate (FPR) of 0.16.
• The traditional Chan-Vese model achieved an average F-measure of 0.52 and an FPR of 0.32.
• The proposed model demonstrated superior performance over the Chan-Vese model in efficiency, universality, and lesion detection.

Conclusions:

• The Hue-texture-embedded region-based model provides superior segmentation for conditions like chronic gastritis, intestinal metaplasia, and early gastric cancer compared to traditional active contour methods.
• Future work aims to expand the model's applicability to segment other lesions, such as intramucosal cancer.
• Successful implementation will pave the way for a fully automatic computer-assisted diagnosis system for ME-NBI images.