1

Automatic measurement of D-score in human endometrium images Javier Rojas

Abstract— This document presents a method to identify an reconstruct endometrium glands in images of samples of endometrium tissue, using non-rigid registration.

I. I NTRODUCTION The endometrium hyperplasia is a pathology which is correlated to the presence of adenocarcinoma, and its presence is used to estimate the likelihood of developing this kind of cancer. To diagnose endometrium hyperplasia the pathologist used different methods, such as verbal descriptions [1], making thus to this process something difficult of reproduce and validate. In 1985 an advance was presented with the introduction of the WHO classification scheme [1], that allowed some standardization of the diagnosis process. This scheme is based on the evaluation of qualitative characteristics of the analyzed samples, such as the nuclear atipicity and the nuclear complexity of the glands. Despite the advances made with this methodology over the traditional diagnosis process, the WHO scheme has sensibility problems [1]; there are several diagnosis that according to the WHO schema are severe and that finally don’t develop cancer. The medical community agrees about considering the WHO scheme insufficient to do a proper diagnosis of endometrium hyperplasia [2], [3]: several studies have found that this kind of diagnosis is

hard to reproduce, thing that makes this scheme less reliable and less sensitive. The indicator known as D-score [4] has shown a major sensitivity that the evaluation of qualitative characteristics made by the WHO classification [2]. This indicator considers qualitative and quantitative characteristics of the endometrium samples used for the diagnosis, and is earning space as an standard method in this task. The D-score is defined as:

D-score = 0.6229 + 0.0439 × (porcentaje de volumen del estroma

−3.9934 × ln(desviaci´on est´andar de los ejes nucleares m´as cortos −0.1592(densidad de la superficie superior de las gl´andulas) (1)

II. R ELATED WORK Una de las razones m´as importantes para seguir haciendo este an´alisis de manera manual es que las im´agenes presentan una gran cantidad de ruido, variaciones de color, de forma, junto con discontinuidades en las formas buscadas, ver figura 1(b) on the following page. Estas dos caracter´ısticas —la variabilidad de formas y la discontinuidad en los segmentos de las gl´andulas— hacen de la medici´on del D-score un problema interesante de procesamiento de im´agenes; el reconocimiento y reconstrucci´on de las gl´andulas se pueden plantear de manera m´as formal en el a´ rea como detecci´on de formas cerradas y la conexi´on de bordes (edge linking).

2

(a)

(b)

(c) Fig. 1.

Im´agenes t´ıpicas de muestras de endometrio

Varias aproximaciones autom´aticas en el a´ rea para analizar im´agenes de microscopia o histolog´ıa [5]–[8] no son adecuadas para este problema debido a suposiciones muy fuertes sobre el tipo de imagen y las formas en ella, tales como que las formas se han conservado, que pueden modelarse matem´aticamente, etc. T´ecnicas basadas en la topolog´ıa de la imagen, como la usada en [9], permite encontrar de manera directa las regiones cerradas —las gl´andulas— en la misma. Sin embargo, este enfoque no es aplicable directamente debido a la presunci´on de que los objetos buscados en la imagen no presentan discontinuidades en sus bordes; en otras palabras, que los contornos

son f´acilmente identificables y se mantienen cerrados. Nattkemper resume en [10] varias t´ecnicas para detectar c´elulas en im´agenes de histolog´ıa. Una aproximaci´on bastante usada es la transformada Hough [11], que usa una definici´on matem´atica dada de antemano que describe el objeto buscado (como una l´ınea o una elipse, por ejemplo) para ubicar los objetos de inter´es en la imagen, estimando los par´ametros que se ajustan mejor a cada forma encontrada. Aunque la transformada Hough permite resolver el problema de la discontinuidad en los segmentos que definen la gl´andula tiene varias desventajas: s´olo permite reconocer formas definidas de antemano, e´ stas formas deben poder ser descritas matem´aticamente, y es muy lento. Hai-Shan [5] plantea otro esquema para detectar formas mediante estimaci´on de par´ametros, similar a la transformada Hough, y muestra su aplicaci´on en la detecci´on de contornos de c´elulas, que desafortunadamente comparte los mismos problemas de la transformada. El problema de conexi´on de segmentos discontinuos en im´agenes binarias tambi´en ha sido examinado desde otras perspectivas; Zhu [12] presenta una estrategia basada en funciones de potencial, en la que los extremos de los segmentos se atraen mutuamente, con una fuerza inversamente proporcional a la distancia que los separa. Un problema con la t´ecnica es que la distancia entre segmentos no es suficiente para discriminar los segmentos de cada gl´andula, ya que e´ stas suelen estar agrupadas densamente en las im´agenes (ver figura 1(b)). Shih [13] propone el uso de operaciones de morfolog´ıa matem´atica para modificar los extremos de los segmentos de manera iterativa hasta conectar los bordes de los mismos. De-

3

bido a la naturaleza de estas operaciones, se depende de la orientaci´on inicial de los extremos de los segmentos para que el m´etodo funcione adecuadamente.

the person using the training application selects which pixels belong to which categories. All the pixels selected are used to train the multilayer perceptron; the testing of the classifier is made interactively with other test images.

III. M ETHODOLOGY A. Image preprocessing and segmentation

B. Primitive extraction

Thanks to the staining procedure applied to the samples it is possible to identify visually three different kinds of tissue within the image, see Fig. 1 on the previous page 1) The border of the endometrium glands, with light brown color. 2) The nuclei of the glands, in white color. 3) The background of the sample, with pink color. The identification of every tissue type is made through a classification procedure based on the RGB colors in the image, using a machine learning classifier. 1) Image preprocessing: To correct the lighting and staining differences in the processed images —caused by environmental conditions, the stain medium application, and other factors [14]— is necessary a preprocessing stage that equalizes the analyzed image. This process allows to train the classifier with some typical images, and then equalize the histogram of the analyzed image to match more closely the histogram of the images used for training, and hence achieving better performance when processing the image with the classifier. TODO: describe how does it work 2) Image segmentation: A multilayer perceptron is used to segment the image; the segmentation quality of this classifier has shown experimentally good enough to identify the gland borders in the analyzed images. TODO: describe the architecture of the NN. The training data for the classifier is selected manually from a subset of the test images, and

After the segmentation process we have a binary image, with the interest objects —the gland borders— in black, and the glands nuclei and background in white. A topological tree of the image’s objects is built and used to improve the results obtained from the segmentation. The whole tree is explored to discard topological nonsenses and to identify the different kind of resulting objects, such as closed glands, or different gland segments. Any region too small (less than 40 pixels) was discarded, and made part of its enclosing region. Besides, any region below the third level of the tree was discarded, and recursively made part of the white region enclosing it. This is a topological impossibility, since it is assumed that the image contains only closed shapes (white-black-white paths in the topological tree), or disjoint segments (white-black paths). In this stage is possible to identify glands that are too close from each other, and as result of this. C. Primitive grouping The idea behind this stage is to perform registration considering only primitives that are likely to form a particular closed gland. To do this, several subsets of primitives are built, and the registration process is performed over each one of them. To build these, several criterion are considered, such as euclidean distance

4

between them, their shape, an approximated measure of convexity, and their orientation. However D. Shape registration To be made E. Gland reconstruction And this F. Measurement of the D-score And this IV. R ESULTS I’ll be talking only about D-score stuff and registration here, not segmentation. R EFERENCES [1] R. Kurman and P. K. H. Norris, “The behavior of endometrial hyperplasia. A long-term study of untreated hyperplasia in 170 patients,” Cancer, vol. 56, pp. 403–412, 1985. [2] A. Ørbo, J. P. A. Baak, I. Kleivan, S. Lysne, P. S. Prytz, M. A. M. Broeckaert, A. Slappendel, and H. J. Tichelaar, “Computerised morphometrical analysis in endometrial hyperplasia for the prediction of cancer development. A long term retrospective study from northern Norway,” Journal of Clinical Pathology, vol. 53, pp. 697–703, 2000. [3] C. Dunton, J. Baak, J. Pallazo, and etal, “Use of computerized morphometric analyses of endometrial hyperplasias in the prediction of coexistent cancer,” Am J Obstet Gynecol, vol. 174, pp. 1518– 1521, 1996. [4] J. Baak, J. Nauta, E. Wisse-Brekelmans, and et al, “Architectural and nuclear morphometrical features together are more important than prognosticators in endometrial hyperplasia than nuclear features alone,” Journal of Pathology, vol. 154, pp. 335–341, 1988. [5] H.-S. Wu, J. Barba, and J. Gil, “A parametric fitting algorithm for segmentation of cell images,” IEEE Transactions on Biomedical Engineering, vol. 45, pp. 400–407, 1998.

[6] A. Jalba, M. Wilkinson, and J. Roerdink, “Automatic segmentation of diatom images for classification.” Microscopy Research and Technique, vol. 65, pp. 72–85, 2004. [7] J. Lindblad, C. W¨ahlby, E. Bengtsson, and A. Zaltsman, “Image analysis for automatic segmentation of cytoplasms and classification of rac1 activation,” Cytometry, vol. 53, pp. 22–33, 2003. [8] F. J. Sanchez-Marin, “Automatic segmentation of contours of corneal cells.” Comput Biol Med., vol. 29, pp. 243–258, 1999. [9] O. Cuisenaire, E. Romero, C. Veraart, and B. M. M. Macq, “Automatic segmentation and measurement of axons in microscopic images,” in Image Processing, 1999. [10] T. W. Nattkemper, “Automatic segmentation of digital micrographs: A survey,” medinfo, 2004. [11] P. Hough, “Method and means for recognizing complex patterns,” U.S. patent 3069654, Dec. 18, 1962. [12] Q. Zhu, M. Payne, and V. Riordan, “Edge linking by a directional potential function (dpf),” Image and Vision Computing, vol. 14, pp. 59–70, 1996. [13] F. Y. Shih and V. Gaddipati, “General sweep mathematical morphology,” Pattern Recognition, vol. 36, pp. 1489–1500, 2003. [14] C. Demir and B. Yener, “Automated cancer diagnosis based on histopathological images: a systematic survey,” Rensselaer Polytechnic Institute, Department Of Computer Science, Tech. Rep., 2005.