A Multi-Stage Recognition System to Detect Different Types of Abnormality in Capsule Endoscope Images

Beitrag in einer Fachzeitschrift
(Originalarbeit)


Details zur Publikation

Autor(en): Miaou SG, Chang FL, Timotius I, Huang HC, Su JL, Liao RS, Lin TY
Zeitschrift: Journal of Medical and Biological Engineering
Verlag: Walter H Chang
Jahr der Veröffentlichung: 2009
Band: 29
Heftnummer: 3
Seitenbereich: 114 - 121
ISSN: 1609-0985
Sprache: Englisch


Abstract


The capsule endoscope is a state-of-the-art tool to detect abnormal problems of the small intestines, such as chyme blockage, suspected blood indicator (SBI), and white spots (ulcer). However, each examination using a capsule endoscope produces several tens of thousands images, and it is time-consuming for a physician to examine all the images. This paper proposes an automatic recognition system to identify suspected capsule endoscope images to save image viewing time. The system is basically a four-stage classifier. The first stage uses the hue, saturation, and intensity (HSI) color model to find the images with large yellow-green abnormal areas (could be chyme-blocked). For those not selected in the first stage, the second stage uses fuzzy c-means clustering analysis to further recognize the images with large abnormal areas (could be SBI). Most of the images encountered at the third stage have either large normal and uniform areas, or small abnormal areas. Thus, this stage attempts to exclude uniform images which are likely to be normal. Finally, the last stage uses a back-propagation neural network to detect the images with small abnormal areas (could be white spots). Experimental results showed that this cascaded classification system could perform much better than its individual stages or some combinations of the stages. The overall recognition accuracy is about 89%, resulting in unavoidable misclassified images. However, the detection sensitivity for abnormal images by the system was as high as 97.7%. Furthermore, since abnormality of the small intestines usually appears in a group of consecutive images, it would be overlooked only when every abnormal image in the group was misclassified. Thus, as an abnormality screening tool for physicians, the proposed system is valuable.



FAU-Autoren / FAU-Herausgeber

Timotius, Ivanna
Lehrstuhl für Informatik 5 (Mustererkennung)


Zitierweisen

APA:
Miaou, S.-G., Chang, F.-L., Timotius, I., Huang, H.-C., Su, J.-L., Liao, R.-S., & Lin, T.-Y. (2009). A Multi-Stage Recognition System to Detect Different Types of Abnormality in Capsule Endoscope Images. Journal of Medical and Biological Engineering, 29(3), 114 - 121.

MLA:
Miaou, Shaou-Gang, et al. "A Multi-Stage Recognition System to Detect Different Types of Abnormality in Capsule Endoscope Images." Journal of Medical and Biological Engineering 29.3 (2009): 114 - 121.

BibTeX: 

Zuletzt aktualisiert 2018-07-08 um 00:47