Analysis of mammogram classification using a wavelet transform decomposition

https://doi.org/10.1016/S0167-8655(02)00221-0Get rights and content

Abstract

In order to fully achieve automated mammogram analysis one has to tackle two problems: classification of radial, circumscribed, microcalcifications, and normal samples; and classification of benign, malign, and normal ones. How to extract and select the best features from the images for classification is a very difficult task, since all of those classes are basically irregular textures with a wide visual variety inside each class. Besides there is a lack of tested solutions for these problems in the literature. In this paper we propose to construct and evaluate a supervised classifier for these two problems, by transforming the data of the images in a wavelet basis, and then using special sets of the coefficients as the features tailored towards separating each of those classes. We have realized that this is a suitable solution worth further exploration. For the experiments we have used samples of images labeled by physicians. Results shown are very promising, and the paper describes possible lines for future directions.

Introduction

Feature selection and classification are the cornerstone processes of a pattern recognition problem. In the case of image data one has to decide whether arrangements of spatial data (i.e., pixels directly) can be used as elements of features, or if a transformation of the pixels to a different space can uncorrelate the meaningful information needed to separate the data into the classes desired. For image analysis problems a texture is a challenging feature to recognize, since it is often not a regular pattern and it is very dependent on scale.

Breast cancer is a major occurrence of cancer in women over 40 years old nowadays all over the world. An early diagnostic is very important, and one common method of diagnosis is by using a mammogram, which is basically an X-ray of the breast region taken in a special condition. Fig. 1 shows a typical mammogram. From the image a trained physician screens it searching for microcalcifications, and masses of various forms. If found, these artefacts on the image could be a sign for the presence of a benign or malign tumor. This is yet a challenging and unsolved problem for typical pattern recognition approaches, mainly because the microcalcifications and masses appear as almost free shapes, and there are vessels and muscles which are more or less prominent in the images depending on the patient.

We propose in this paper to construct and evaluate a supervised classifier for mammograms using a wavelet transform decomposition. The experiments performed show that a successful classification can be achieved, even when we consider the two main problems: (1) classification between normal, benign, and malign areas; (2) classification between normal, microcalcifications, radial or spiculated, and circumscribed areas. There is a lack of tested solutions for these problems in the literature, and this paper aims to contribute to that. Section 2 shows the images of typical mammograms and its target classes, along with a revision of literature on mammograms classification. Section 3 defines the problem in terms of a pattern recognition framework and presents a proposed multiresolution approach for its solution. Section 4 shows experiments on images taken from MIAS.1 Section 5 gives conclusions and points to future extensions.

Section snippets

Mammograms

The acquisition of a mammogram is done by compressing the breast of the patient between two acrylic plates for a few seconds when the X-ray is emitted. A typical mammogram is an intensity image with gray levels, showing the levels of contrast inside the breast which characterize normal tissue, vessels, different masses of calcification, and noise. A typical mammogram is shown in Fig. 1.

Some calcifications can be grouped in classes according to their similar geometrical properties. They are

Texture analysis and a pattern recognition framework using a multiresolution approach

In a general way texture can be characterized as the space distribution of the gray levels in a neighborhood. Jain et al. (1995) define texture as the variation pattern of the gray levels in a certain area. Texture is a feature that can not be defined for a point, and the resolution at which an image is observed determines the scale at which the texture is perceived. So, texture is a confusion measurement that depends mainly on the scale which the data are observed. There are textures with

Experiments

Experiments were accomplished for the two problems: the geometric property of the tumor, and its nature. The first set of experiments took into consideration the geometric property of the tumor, considering four classes: radial lesions, circumscribed lesions, microcalcifications and normal areas.

The images used in this set of experiments are shown by class. Some noisy images were obtained from original ones and used for testing, namely ndbXXX, rdbXXX and sdbXXX.

The noisy images were obtained by

Conclusions

Having a complete automated mammogram analysis solution is still something to be achieved. In this paper we have shown a way to construct a solution considering the two main classification problems of mammograms (i.e. nature of tumor and related geometries). We have also evaluated the solution in real data, and the results point to a promising solution to be further explored.

We have proposed in this paper that by using a wavelet transformation of the data, we can devise a pattern recognition

Acknowledgements

This work has been partially supported by PUCPR with a student bursary given to the first author C.B.R. Ferreira.

References (10)

  • GIMP - Mattis, P., Kimball, S., The GNU Image Manipulation Program, released the GNU General Public License (GPL), for...
  • Guimarães, M.S., 1999. Abordagens Difuso-Neurais para Análise de Mamogramas, M.Sc. dissertation (in Portuguese), Centro...
  • Jacobs, C.E., Finkelstein, A., Salesin, D.H., 1995. Fast Multiresolution Image Querying, Proceedings of SIGGRAPH 95,...
  • R Jain et al.

    Machine Vision

    (1995)
  • S.G Mallat

    A theory for multiresolution signal decomposition: the wavelet representation

    IEEE Trans. Pattern Anal. Machine Intelligence

    (1989)
There are more references available in the full text version of this article.

Cited by (98)

View all citing articles on Scopus
View full text