Self-training with unlabeled regions for NBI image recognition

In this paper, we propose a self-training method which uses unlabeled regions in the original images obtained from a colorectal Narrow Band Imaging (NBI) zoom-video endoscope. The proposed method first trims a number of patches from unlabeled regions in the original images and uses them as unlabeled training samples. Classifiers are trained with the available labeled samples, as well as with those unlabeled training samples, using a newly-proposed rejection condition which takes into account the class asymmetry of the NBI images. Experimental results demonstrate that the proposed method improves performance with a statistically significant difference.

1 Introduction

Colorectal cancer has been one of the major causes of cancer death all around the world [1], [2]. An early detection of colorectal cancer through colorectal endoscopy is important and widely used in hospitals as a standard medical procedure. During colonoscopy, the lesions of colorectal tumors on the colon surface are visually inspected by an Narrow Band Imaging (NBI) zoom-videoendoscope. By using the visual appearance of the colorectal tumors in endoscopic images, histological diagnosis is presumed based on classification schemes for NBI magnification findings [3] (Figure 2). A diagnosis by visual inspection, however, is affected by the skill and familiarity of each endoscopists. Hence, a computer-aided system for supporting the visual inspection would be of great help for colonoscopy, due to the large number of images of colorectal tumors which should be classified in a periodical medical examination for detecting cancer in its early stage.