発表論文

2009.06

(コンピュータビジョンとイメージメディア (CVIM) Vol. 2009-CVIM-167)

HAI VU, TOMIO ECHIGO, 佐川 立昌, Yagi Keiko, Shiba Masatsugu, Higuchi Kazuhide, Arakawa Tetsuo

概要

Video capsule endoscopy (VCE) represents a significant advance in examinations of digestive diseases by providing a non-invasive method to view the small bowel. In addition, VCE provides a valuable source for visualizing the intestinal contractions, which are mainly events for intestinal motility assessment. However, the advantages of VCE diagnosis technique are facing with the time consuming for reading video sequence as well as challenging to detect the intestinal contractions. In this paper, we present our works to approach these motivations through techniques of VCE analysis. VCE interpretations could be implemented by analyzing spatial and temporal features. First, several image features such as color, edges, and motion displacement are extracted. Then their temporal analyses are presented in several ways to adapt with different tasks. Two applications utilizing this framework are developed. In the first application, we propose a new method to reduce diagnostic time under the constraint that all original images should be displayed to an examining doctor without skipping frames. To realize such a system, delay time for drawing images between frames is controlled in adaptive rate, according to the states of capturing images. Several techniques for the state classification, delay time calculation, and log-based analysis are deployed in this application. In the second application, we develop a three-stage procedure for the intestinal contraction detections. Based on the characteristics of contractile patterns, the possible contractions can be investigated using essential images features extracted from VCE such as changes in edge of the intestinal folds and by evaluating similarities features in consecutive frames. Then true contractions are determined through spatial analysis of directional information. To exclude as many non-contractions as possible, we consider about information of contractions frequencies along capsule transit time. Both the quality and quantity indices are analyzed in experiments for performance evaluations.Video capsule endoscopy (VCE) represents a significant advance in examinations of digestive diseases by providing a non-invasive method to view the small bowel. In addition, VCE provides a valuable source for visualizing the intestinal contractions, which are mainly events for intestinal motility assessment. However, the advantages of VCE diagnosis technique are facing with the time consuming for reading video sequence as well as challenging to detect the intestinal contractions. In this paper, we present our works to approach these motivations through techniques of VCE analysis. VCE interpretations could be implemented by analyzing spatial and temporal features. First, several image features such as color, edges, and motion displacement are extracted. Then their temporal analyses are presented in several ways to adapt with different tasks. Two applications utilizing this framework are developed. In the first application, we propose a new method to reduce diagnostic time under the constraint that all original images should be displayed to an examining doctor without skipping frames. To realize such a system, delay time for drawing images between frames is controlled in adaptive rate, according to the states of capturing images. Several techniques for the state classification, delay time calculation, and log-based analysis are deployed in this application. In the second application, we develop a three-stage procedure for the intestinal contraction detections. Based on the characteristics of contractile patterns, the possible contractions can be investigated using essential images features extracted from VCE such as changes in edge of the intestinal folds and by evaluating similarities features in consecutive frames. Then true contractions are determined through spatial analysis of directional information. To exclude as many non-contractions as possible, we consider about information of contractions frequencies along capsule transit time. Both the quality and quantity indices are analyzed in experiments for performance evaluations.