Unsupervised abnormality detection using saliency and Retinex based color enhancement.

Area of Science:

• Medical Imaging
• Computer Vision
• Gastroenterology

Background:

• Capsule endoscopy generates vast visual data, necessitating efficient automated methods for abnormality detection.
• Pre-processing techniques like color enhancement can improve image quality and diagnostic performance.

Purpose of the Study:

• To propose and evaluate a two-stage automated algorithm for detecting abnormalities in capsule endoscopic images.
• To assess the impact of adaptive color enhancement on the performance of abnormality detection.

Main Methods:

• A two-stage algorithm was developed: 1) adaptive color enhancement using Retinex theory, and 2) salient region detection for identifying clinically significant areas.
• The algorithm was tested on a diverse dataset of capsule endoscopic images exhibiting various pathologies.

Main Results:

• The proposed algorithm successfully detected a significant percentage of abnormal regions.
• Color enhancement demonstrably improved the performance of the abnormality detection system.
• The system achieved a sensitivity of 97.33% and a specificity of 79%, outperforming existing methods.

Conclusions:

• The developed two-stage automated method, incorporating adaptive color enhancement, is effective for abnormality detection in capsule endoscopy.
• This approach offers a promising solution for reducing the screening burden and improving diagnostic accuracy in capsule endoscopy procedures.