Feature extraction for abnormality detection in capsule endoscopy images

Capsule endoscopy is a non-invasive method for diagnosing gastrointestinal diseases. This new technology has many advantages over conventional endoscopy. However, investigating endoscopic video frames in search of diseases is a tedious task for physicians. Hence, a system is required to automatically detect suspicious frames for further medical examination. Different abnormalities may exist in capsule endoscopy images. In this paper, a novel method is proposed to investigate capsule endoscopy images for abnormalities such as bleeding and angiodysplasia lesions. The proposed method identifies potential regions of interest by using an expectation maximization based image segmentation algorithm, and extracts features from them using a combination of color histogram analysis and statistical features to classify frames into normal and abnormal classes. The results show that the proposed method can distinguish the two objective classes with an approximate precision and recall of 96.5% and 95.9%, respectively.