Abnormal pattern detection in Wireless Capsule Endoscopy images using nonlinear analysis in RGB color space

In recent years, an innovative method has been developed for the non-invasive observation of the gastrointestinal tract (GT), namely Wireless Capsule Endoscopy (WCE). WCE especially enables a detailed inspection of the entire small bowel and identification of its clinical lesions. However, the foremost disadvantage of this technological breakthrough is the time consuming task of reviewing the vast amount of images produced. To address this, a novel technique for distinguishing pathogenic endoscopic images related to ulcer, the most common disease of GT, is presented here. Towards this direction, the Bidimensional Ensemble Empirical Mode Decomposition was applied to RGB color images of the small bowel acquired by a WCE system in order to extract their Intrinsic Mode Functions (IMFs). The IMFs reveal differences in structure from their finest to their coarsest scale, providing a new analysis domain. Additionally, lacunarity analysis was employed as a method to quantify and extract the texture patterns of the ulcer regions and the normal mucosa, respectively, in order to discriminate the abnormal from the normal images. Experimental results demonstrated promising classification accuracy (>95%), exhibiting a high potential towards WCE-based analysis.