Feature Selection for Bleeding Detection in Capsule Endoscopy Images using Genetic Algorithm

Wireless Capsule Endoscopy (WCE) is an advantageous tool for diagnosing gastrointestinal disorders. An automated analysis method is required to detect abnormality because a large number of images is produced along WCE journey through the patient's digestive tract. In this paper, a computer-aided method is proposed to detect bleeding frames. Bleeding is the most common abnormality in WCE images. This abnormality is distinguishable by using its color features. Hence, in this method, at first, different statistical features are extracted from color channels in different color spaces. Then, a feature selection method based on genetic algorithm is proposed to select a subset of the appropriate features. The experimental results show that the proposed feature selection method increases the detection accuracy and reduces the size of initial features set by almost half. The achieved accuracy, recall and precision of the proposed method are 97.58, 96.76 and 98.29, respectively.