A Hessian-Based Technique for Specular Reflection Detection and Inpainting in Colonoscopy Images

Area of Science:

• Medical Imaging
• Computer Vision
• Artificial Intelligence

Background:

• AI-based algorithms are increasingly used for disease detection in endoscopy, particularly colonoscopy.
• Specular reflections in colonoscopy images often lead to false positives, hindering AI performance.
• Existing methods struggle to effectively address specular reflection artifacts.

Purpose of the Study:

• To develop and evaluate a novel algorithm for restoring specular reflection regions in colonoscopy images.
• To improve the accuracy and reliability of AI-based Computer-Aided Detection (CADx) systems.
• To enhance the visual quality and diagnostic utility of endoscopic images.

Main Methods:

• Conversion of RGB images to HSV color space, focusing on the Saturation (S) component.
• Detection of convex regions using a Hessian-based method to identify specular reflections.
• Precise inpainting and feathering techniques to restore affected regions and blend them seamlessly.

Main Results:

• The algorithm effectively identifies and restores regions with high specular reflection.
• Inpainting and feathering processes ensure aesthetic coherence and accurate restoration.
• Superior performance demonstrated across five colonoscopy datasets and Kvasir dataset images compared to existing methods.

Conclusions:

• The proposed algorithm significantly enhances the performance of AI in colonoscopy by mitigating specular reflection artifacts.
• Accurate restoration of specular reflection regions leads to improved diagnostic accuracy in endoscopic imaging.
• This technique offers a promising solution for improving Computer-Aided Detection (CADx) systems in gastroenterology.