Segmentation and Analysis of Breast Tumors on Ultrasonography
 

 

Miguel Alemán-Flores 1
Patricia Alemán-Flores 2
Luis Álvarez-León 1
M. Belén Esteban-Sánchez 1
Rafael Fuentes-Pavón 2
José M. Santana-Montesdeoca 2

 

(1 (1) Departamento de Informática y Sistemas. Universidad de Las Palmas de Gran Canaria, Spain.
 (2) Sección de Ecografía. Servicio de Radiodiagnóstico. Hospital Universitario Insular de Gran Canaria, Spain.
 

maleman@dis.ulpgc.es          

http://serdis.dis.ulpgc.es/~maleman

   

Introduction

The examination by means of medical imaging can reduce the number of biopsies necessary to discriminate malignant from benignant breast tumors. The main types of breast medical images are mammography and ultrasonography. This work presents a new approach in the application of computer vision techniques to the diagnosis of solid breast nodules in ultrasonography. The studies carried out by specialists have proved the importance of certain features which help determine whether a nodule is benign or malignant. It is quite complex for a human observer to analyze some of these features, but computer vision techniques can make it easier to detect them and generate numerical measurements of their presence and relevance, so that the risk of overlooking a malignant nodule can be reduced.
 
We present a combination of different techniques to obtain an accurate segmentation of a nodule from an inner point. Furthermore, we propose a set of methods to extract objective evidence of the presence or absence of the diagnostic criteria by means of an exhaustive analysis of the general shape of the nodule, the local variations of its contour and the contrast between certain areas. This application aims to detect and measure the relevant features with a high accuracy and be a valuable help for radiologists. The purpose of this work is not  the substitution of the specialists, but the introduction of a precise, objective and invariable point of view which can draw their attention to those areas where each factor has a high probability of being present.
 

Materials and Methods

 

 
We have developed a common framework for the delimitation of the nodules and the further analysis of the diagnostic criteria, so that they can be measured by the computer and the physicians have a set of parameters available to decide whether the biopsy is necessary or not.
 
Segmentation of the nodule
We filter the image using the truncated median filter. By means of a combination of a region-growing algorithm and active contour technique, we obtain a precise delimitation of the nodule from a single inner point [1][2].

 

 

Analysis of the diagnostic criteria

For the analysis of breast ultrasound, we have used the following criteria (from Stavros et al. [3]):
 
Benignity Criteria
1. Hyperechogenicity
2. Ellipsoid shape
3. Two or three gentle lobulations
4. Thin echogenic capsule
 
Malignancy Criteria
5. Hypoechogenicity
6. Taller-than-wide areas
7. Spiculation
8. Angular Margins
9. Microlobulations
10. Acoustic shadowing
11. Microcalcifications
12. Ramifications
 

Results

 

The segmentations of the nodules obtained with the proposed technique are quite similar to those provided manually by the specialists. On the other hand, the detection of the diagnostic criteria is coherent, in most cases, with the appreciation of the specialists. The following figures show the segmentation and the results generated for some criteria in the analysis of several nodules:

 

Conclusion

The possibility of generating objective numerical measures of the presence of the criteria represents a considerable progress in the classification of the nodules. The wide range of features which we have considered and the robustness and accuracy of the measures reflect their usefulness in the discrimination of malignant and benignant nodules. These results show the importance of the introduction of computer vision techniques in the analysis of medical images. We are currently in the process of validating the models which have been proposed and adapting the parameters by comparing our results with  the descriptions provided by the radiologists.

 
[1] Alemán-Flores, M., Alemán-Flores, P., Álvarez-León, L., Santana-Montesdeoca, J.M., Fuentes-Pavón, R., Trujillo-Pino, A.: Computational Techniques for the Support of Breast Tumor Diagnosis on Ultrasound Images. Cuadernos del Instituto Universitario de Ciencias y Tecnologías Cibernéticas. ULPGC, 2003; 27:1-12.
[2] Caselles, V., Kimmel, R., Sapiro, G.: Geodesic Active Contours. In Procs of ICCV’95, 1995; 694-699.
[3] Stavros, A.T., Thickman, D., Rapp, C.L., Dennis, M.A., Parker, S.H., Sisney, G.A.: Solid Breast Nodules: Use of Sonography to Distinguish between Benign and Malignant Lesions, Radiology, 1995; 196:123-134.
 
 
Bildverarbeitung für die Medizin – Heidelberg – Germany – March 2005