D. Roller , C. Lampasona
The objective of this project was to develop and apply image analysis techniques to the study of digital mammograms to measure breast density in a quantitative reliable practical reproducible way, from a three-dimensional point of view. We aimed also to make possible a better analysis of the evolution of the patient based on several time-shifted images. This was achieved through the development of a new standardization for digital mammograms. Thanks to this standardization, many images can be automatically compared.
The statistics of 2004 show that more than 47500 women get breast cancer annually in Germany, 19300 of them at an age under 60 years. Breast cancer represents the most frequent cancer within women. This disease is responsible for 24.4% of all new cases of cancer within women and more than one third (34%) of the new diseases within women under 60 years. The statistics of 2006 show an increase of women suffering of breast cancer (55150), with an average disease age of 62 years. During 2004, 17592 women died of breast cancer in Ger-many.
In the early phase of the disease, it often does not have symptoms perceivable by the woman. The most widespread procedure to detect breast cancer includes clinical examination by a trained professional, self-examination, and breast imaging. There are two kinds of breast imaging tests: 1) screening, which is performed in patients with no symptoms to detect cancer when it is still too small to be felt by a woman or her physician; and 2) diagnostic, which is performed in women who either have a breast complaint or have had an abnormality found during screening.
The conduction of mammography-screening in Germany started in 2005 after many years of preparation, including the implementation of model projects, to test the use of the “European Guidelines for the Quality Assurance of Mammography-Screenings” under the conditions of the German health system. This screening follows a resolution of the Federal Ministry of Health of 1 January 2004.
In general, a mortality reduction of 20% to 30% is expected, which means a significant benefit for the women participating in the screening. The more influencing breast cancer risk factors are: being a woman; being 50 years of age or older; familiar history of breast cancer; having suffered of other malign diseases; early menarche; late pregnancy; childlessness; short lactation; and overweight.
The work of Wolfe showed that women, whose breasts exhibit an enlargement and/or a change of the glandular tissue have a higher breast cancer risk. Other authors found also a connection between increased breast cancer risk and Wolfe’s categories.
Medical image analysis can help to find abnormal changes in the breast tissues from digital mammograms. Increased mammographic breast density is such a change and a factor that influences the risk of becoming affected with breast cancer. On the other hand, unlike other risk factors, it could be quantified. The estimation of breast density could then help to find high-risk populations and to take preventive decisions.
In the early 1980s, the first attempts to develop quantitative methods to measure breast density began. These approaches were based on visual assessment and planimetry. Breast density was meant as a two-dimensional mammographic density, the fraction of the image occupied by glandular tissue. Many authors have already developed methods to estimate the percentage of glandular tissue but from a two-dimensional point of view. Considering the measurement of breast density as a two-dimensional problem, takes into account only a part of a three-dimensional phenomenon. For example, women with large breast may have a great amount of glandular tissue but low breast density. On the other hand, women with small breast and less glandular tissue may have a higher breast density.
New methods were developed, other methods were optimized and applied and some techniques from other areas were adapted for our purposes. These methods are the following:
- Segmentation of the image into background and breast;
- Pectoral muscle segmentation in the mediolateral-oblique images;
- Interpretation of pixels gray levels to determine which combinations of tissue they represent;
- Detection of the breast border;
- Detection of calcification;
- Conversion of the images into a standard for comparison;
- Registration and comparison of the images;
- 2D-visualization of the results and
- 3D-visualization of the results.
The project is supported with a graduated scholarship of the Heinrich-Böll-Stiftung.
|
UZ) | © Universität Stuttgart | Legal Notice