Endoscopy video analysis algorithms and
their independence of rotation, brightness,
contrast, color and blur

Jan Cychnerski

Endoscopy video analysis algorithms and their independence of rotation, brightness, contrast, color and blur

Jan Cychnerski
Department of Computer Architecture, Gda´nsk University of Technology, Poland

Related article at Pubmed, Scholar Google

Visit for more related articles at International Journal of Innovative Research in Computer and Communication Engineering

Abstract

The article presents selected image analysis algorithms for endoscopy videos. Mathematical methods that are part of these algorithms are described, and authors’ claims about the characteristics of these algorithms, such as the independence of rotation, brightness, contrast, etc. are mentioned. Using the common test on the real endoscopic image database and a set of image transformations, the validity of these claims was checked and compared between algorithms. Many of the results seem to differ from the declaration of the authors, sometimes even strongly denying them. In addition, some algorithms were found extraordinary sensitive to blurring of the images, which indicates the possibility of using them for the detection of blurry frames, not just diseases.

Keywords

endoscopy, video analysis, independence, rotation, brightness, contrast, color, blur

INTRODUCTION

In the past several years, endoscopic movie analysis algorithms (obtained by gastroscopy, colonoscopy or capsule endoscopy) were gaining popularity. These algorithms are used to recognize informative and non-informative frames, and various diseases or healthy tissues. Unfortunately, the algorithms described in the literature do not have any larger comparative tests, which makes a comparison between them almost impossible. In addition, the authors of mathematical methods used in these algorithms do not provide evidence for their claims (e.g. on the algorithms’ independence of transformations such as rotation or brightness change). [1]

This article focuses on a comparison of selected endoscopic image analysis algorithms and mathematical tools used in them. The algorithms were tested with a special comparator to check if their authors’ claims about the independence of image transformations like rotation, brightness, contrast, blur and color change are trustworthy.

TEST PROCEDURE

All tests were carried out on a common database of real colonoscopy videos [2]. For the tests, 100 random images from the database were selected. The algorithms’ classification part was removed, leaving only the kernel of the algorithms - the calculation of the feature vectors. In the next step, every feature was normalized so that the average (calculated over all the images in the database) was equal to 0 and standard deviation to 1.

Only the normalized feature vectors were analyzed in the article (ignoring the classification component, such as neural networks or support vector machines). In the literature it is often claimed that the algorithms are dependent or independent of the various image transformations. For comparison and test purposes, 5 popular transformations that occur naturally in endoscopy were selected, as in table I.

A. Comparison measure

In the first place, feature vectors from original images were compared with each other using the metrics described below.

Let F(a) be a feature vector of image a for algorithm F, and fi(a) be its i-th normalized value:

To obtain a common comparing base for algorithm F, feature vectors of all original images (a; b; c; :::) were compared with each other. Then, the average base feature difference was calculated as follows::

The difference between the feature vector between original and transformed images was defined analogously, with additional normalization in regard of typical not-transformed image feature vector differences:

MATHEMATICAL OPERATIONS

In this section, mathematical operations used in the analyzed algorithms are described. Most of them focus on the spatial features of the images, brightness changes and edge detection. This is due to the fact that these features are similar to human vision methods of seeing objects, which allows humans to recognize textural anomalies in the image. Some transformations, however, focus more on the color features, as they are the second most characteristic phenomenon differencing healthy tissue from cancer (e.g. shades of gray, black or bright red are found almost only in cancerous tissue).

Most of the algorithms include also some form of statistical analysis of the characteristics of the image. Depending on the analyzed image features, this allows to reduce the number of dimensions of the resulting feature vector or make it independent from the scale, rotation, or change of contrast or brightness of the image.

Some of the transformations are commonly known in the field of image analysis (e.g. Gabor filters or discrete wavelet transforms), but some are designed specifically for endoscopic (e.g. AHT, NTU).

Table III provides information about the chosen algorithms for the analysis of endoscopy videos provided by the authors (rot. = rotation, sca. = scale, bri. = brightness, con. = contrast).

TEST RESULTS

The following figures present the results of tests carried out. Figure 1 shows the differences between the untransformed images, i.e. D0 and 0. Figures 2 – 6 present: differences after brightness change Dbrightness, after contrast change Dcontrast, after color change Dcolor, after rotation Drotation and after blurring Dblur.

CONCLUSION

The article presents mathematical tools used in the gastrointestinal endoscopic video analysis algorithms. The tests show that algorithms’ authors’ claims about some characteristics of their algorithms about independence from the transformations such as rotation, brightness change, etc. does not always comply with the practical results. Moreover, many of the algorithms are surprisingly sensitive to some transformations – in such cases the difference between the transformed images can be greater than that between the other images. This phenomenon questions the usefulness of these algorithms in provided by the authors applications. However, it is worth noting that some of the algorithm are remarkably sensitive to image blur – this fact indicates these algorithms may perform well in the task of blurry frames recognition, which is also useful in systems supporting digestive system diagnosis.

Tables at a glance


Table 1	Table 2	Table 3

Figures at a glance


Figure 1	Figure 2	Figure 3


Figure 4	Figure 5	Figure 6

References

M. Liedlgruber, A. Uhl, Computer-aided decision support systems for endoscopy in the gastrointesti-nal tract: a review., IEEE reviews in biomedical engineering, vol. 4, pp. 73–88, 2011.

J. Cychnerski, A. Brzeski, A. Blokus, T. Dziubich, and M. J?edrzejewski, Konstrukcja bazy danych dla systemu wspomagania diagnostyki chorób przewodu pokarmowego. In: Studia Informatica, 2012, vol. 33, no. 1. [in polish]

Riaz F., Silva F.B., Dinis Ribeiro M., Coimbra M.T., Invariant Gabor Texture Descriptors for Classification of Gastroenterology Images, IEEE Transactions on Biomedical Engineering, vol. 59, no. 10, Pa´zdziernik 2012

Riaz F., Hassan A., Rehman S., Qamar U., Texture Classification Using Rotation- and Scale-Invariant Gabor Texture Features, IEEE Signal Processing Letters, vol. 20, no. 6, 2013. Manjunath B.S., Ma W.Y., Texture Features for Browsing and Retrieval of Image Data, IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 18, no. 8, 1996

P. Doro?zy´nski, T. Dziubich, Ocena mo?zliwo´sci zautomatyzowanej analizy obrazów z bada´n endoskopowych do wspomagania diagnostyki gastropatii wrotnej. Materia?y konferencyjne ICT Young 2012, p. 341-348 [in polish]

Ojala T., Pietikainen M., Maenpaa T., Multiresolution Gray-Scale and Rotation Invariant Texture Classification with Local Binary Patterns, IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 24, no. 7, 2002

J. Cychnerski, P. Doro?zy´nski, T. Dziubich. An algorithm for portal hypertensive gastropathy recognition on the endoscopic recordings. Information Systems Architecture and Technology, No. 35, 2014.

Li B., Meng M. Q.-H., Tumor Recognition in Wireless Capsule Endoscopy Images Using Textural Features and SVM, IEEE Transactions On Information Technology In Biomedicine, vol. 16, no. 3, 2012

Poh Chee Khun, Zhang Zhuo, Liang Zi Yang, Li Liyuan, Liu Jiang, Feature Selection and Classification for WirelessCapsule Endoscopic Frames, Biomedical and Pharmaceutical Engineering,. ICBPE ’09. International Conference, 2009

Häfner M., Liedlgruber M., Wrba F., Uhl A., Vécsei A., Color treatment in endoscopic image classification using multi-scale local color vector patterns, Medical Image Analysis, 16(1): 75–86, January 2012

Liao S., Zhu X., Lei Z., Zhang L., Li S. , Learning multi-scale block local binary patterns for face recognition, Advances in Biometrics;pp. 828–837, 2007

Häfner M., Gangl A., Liedlgruber M., Uhl A., Vecsei A., Wrba F., Pit Pattern Classification using Extended LocalBinary Patterns,In: Proceedings of the 9th International Conference on Information Technology and Applications in Biomedicine (ITAB’09), Larnaca,Cyprus., 2009.

A. Brzeski, J. Cychnerski, Rozpoznawanie chorób uk?adu pokarmowego z wykorzystaniem technik sztucznej inteligencji, in: ZeszytyNaukowe Wydzia?u Elektroniki, Telekomunikacji i Informatyki Politechniki Gda´nskiej., 2011, no. 9, pp. 395–400. [in polish]

G.D. Magoulas, V.P. Plagianakos, oraz M.N. Vrahatis. Neural networkbased colonoscopic diagnosis using on-line learning and differential evolution. Applied Soft Computing, 2004

G.D. Magoulas. Neuronal networks and textural descriptors for automated tissue classification in endoscopy. Oncology Reports, 15, 2006.

V.S. Kodogiannis oraz M. Boulougoura. An adaptive neurofuzzy approach for the diagnosis in wireless capsule endoscopy imaging. International Journal of Information Technology, 13(1), 2007

T. Ojala, M. Pietikäinen, and D. Harwood (1994), "Performance evaluation of texture measures with classification based on Kullback discrimination of distributions", Proceedings of the 12th IAPR International Conference on Pattern Recognition (ICPR 1994), vol. 1, pp. 582 - 585.

Liao S., Zhu X., Lei Z., Zhang L., Li S. , Learning multi-scale block local binary patterns for face recognition, Advances in Biometrics; pp. 828–837, 2007

Canny, J., A Computational Approach To Edge Detection, IEEE Trans. Pattern Analysis and Machine Intelligence, 8(6):679–698, 1986.

R.M. Haralick, K. Shanmugam, oraz I. Dinstein. Textural features for image classification. IEEE Transactions on systems, man, and cybernetics, November 1973

Wu P., Manjunath B.S., Newsam S., Shin H.D., A texture descriptor for browsing and similarity retrieval, Elsevier, Signal Processing: Image Community, vol. 16, no. 1, 2000

Kamarainen J.-K., Kyrki V., Kälviäinen H., Invariance Properties of Gabor Filter-Based Features-Overview and Applications, IEEE Transactions on Image Processing, vol. 15, no. 5, Maj 2006