A decision fusion strategy for polyp detection in capsule endoscopy

Wireless capsule endoscopy (CE) is now routinely used for non-invasive diagnosis of small bowel diseases. But, it still requires manual assessment of the approximately 50,000 study images. Literature has recently investigated automated methods to detect and analyze various anomalies in CE images to improve reading efficiency and reduce variability. We propose such a computer aided diagnosis (CAD) approach to detect small bowel polyps. For supervised classification of polyps, we investigated fusing multiple statistical classifiers based on color, texture and edge features. The combined boosted classifier when evaluated using 1200 CE images outperformed all individual classifiers and achieved a ~90% classification accuracy.