Biomedical Signal Processing and Control 56 (2020) 101668 Contents lists available at ScienceDirect Biomedical Signal Processing and Control journal homepage: www.elsevier.com/locate/bspc Color-based template selection for detection of gastric abnormalities in video endoscopy Hussam Ali a , Muhammad Sharif a , Mussarat Yasmin a , Mubashir Husain Rehmani b a Department of Computer Science, COMSATS University Islamabad, Wah Campus, Pakistan b Department of Computer Science, Cork Institute of Technology (CIT), Ireland a r t i c l e i n f o Article history: Received 8 July 2018 Received in revised form 16 July 2019 Accepted 17 August 2019 Keywords: Abnormality detection Computer aided diagnosis (CAD) Color histogram Segmentation Endoscopy Support vector machine (SVM) a b s t r a c t Computer-aided diagnosis of gastric diseases from endoscopy frames is an important task. It facilitates both the patient and gastroenterologist in terms of time, money and most important health. Colors are the basic visual features of endoscopic images and also provide clues about abnormal regions in endoscopy frames. A variety of color spaces available for representation of color frames. However, we are not certain about which color space is more suitable for representing color features of gastric images. This paper presents a comparison of color features in different color spaces for detection of abnormal areas in chro- moendoscopy (CH) frames. In addition, the CH images are segmented by using an existing color-difference based segmentation method Delta E (E). A framework for automatic segmentation is presented for endoscopy images by selecting a template image in E by using trained models. For classification, colors features are also merged with texture descriptors. The support vector machine (SVM) classifier is trained on color features and also the hybrid color combined texture characteristics. Then the trained classifier is used to group CH frames into abnormal and normal classes. E with manual template selection has achieved 57.44% accuracy and 56.88% accuracy with the automated process. Moreover, the suggested method achieves 86.6% accuracy and 0.91 area under the curve for the classification of gastric lesions. © 2019 Elsevier Ltd. All rights reserved. 1. Introduction Video endoscopy is a normal procedure for screening the gas- trointestinal (GI) tract of a patient for potential abnormalities e.g., cancer, ulcer, and bleeding. An endoscope is not more than a wire with a light source and camera mounted on its distal tip [1]. However, the investigation of the gastrointestinal (GI tract via an endoscope) is a time spending and tedious task for the medi- cal experts. In fact, video endoscopy generates a large number of frames which are then carefully examined by a gastroenterologist [2]. An automated vision-based system can play an important role by searching out malignant frames from all video frames which saves time for the medical experts, especially when medical experts have to screen too many patients [3]. Numerous improvements have been made in the simple white light endoscopy to provide easiness to the gastroenterologist. These advancements highlight the abnormal regions on the mucosal sur- face and to make them more prominent to the medical experts [4]. E-mail addresses: hussamalics@gmail.com (H. Ali), muhammadsharifmalik@yahoo.com (M. Sharif), mussaratabdullah@gmail.com (M. Yasmin), mshrehmani@gmail.com (M.H. Rehmani). The chromoendoscopy (CH) is also an advance form of normal video endoscopy. It is an image-enhanced endoscopy method, in which dyes (methylene blue) is used to stain the mucosal surface and it highlights the mucosal patterns [5]. On the other hand, digital fil- ters and image enhancement techniques are used to manifest the effect of traditional colorant-based chromoendoscopy in the digital (virtual) chromoendoscopy [6]. The extraction of important characteristics from medical images is an important step for detection of abnormalities in a computer- aided diagnosis system. Colors are one of the vital and basic visual descriptors for the detection of gastric abnormalities [7]. Many studies confirm their role in the detection of abnormal frames in an endoscopic sequence [8–10]. The visual description of abnor- malities (e.g., bleeding and ulcer) can be efficiently manifested by color features. Color images are represented in a combination of color channels in specific color spaces and CH frames also consist of three color channels: red, green and blue (RGB). Some well-known color spaces are hue, saturation, & value (HSV), cyan, magenta, yellow, & black (CMYK), XYZ and Lab defined by which is the Inter- national Commission on Illumination (CIE). There are many more color spaces and every color space has its own application-specific advantages. https://doi.org/10.1016/j.bspc.2019.101668 1746-8094/© 2019 Elsevier Ltd. All rights reserved.