Fast Object Detection Using Color Features for Colonoscopy Quality Measurements

Abstract

The effectiveness of colonoscopy depends on the quality of the inspection of the colon. There was no automated measurement method to evaluate the quality of the inspection. To address this, we have been investigating automated post-procedure quality measurement. The limitation of post-processing quality measurement is that quality measurements become available only long after the procedure was over and the patient was released. A better approach is to inform any suboptimal inspection immediately so that the endoscopist can improve the quality of the inspection in real-time during the procedure. Both post-processing and real-time quality measurements require a number of analysis tasks such as detecting a bite-block region as an indicator that a procedure is an upper endoscopy, not colonoscopy, detecting a blood region as an indicator for inflammation or bleeding, and detecting a stool region as an indicator of quality of the colon preparation. Color is the most distinguishable characteristic for differentiation among these object classes and normal pixels. In this paper, we propose a method to detect these object classes using color features. The main idea is to partition very large positive examples of these objects into a number of groups. Each group is called a “positive plane” and is modeled using a convex hull enclosing feature points of that particular group. Comparisons with traditional classifiers such as K-nearest neighbor (K-NN) and Support Vector Machines (SVM) prove the effectiveness of the proposed method in terms of accuracy and execution time that is critical in the targeted real-time quality measurement system.