X-ray mammography is known as an efficient diagnosis method for detection of early stage breast cancers. The mammogram is a radiograph of the breast tissue and an effective non-invasive means of searching for breast cancers. Microcalcifications and masses are two main types of breast abnormalities. For the detection of breast cancer, the mass detection is more difficult than microcalcification detection due to the fact that masses are obscured by normal breast parenchyma. The radiologists have performed mammogram interpretation by visual examination of the films for the abnormalities that can be interpreted as cancerous lesions. However, the observational oversights of radiologists and a large number of mammograms to be examined make readings intensive, cost ineffective, and inaccurate. The computer-aided diagnosis (CAD) will be useful to assist early diagnosis of breast cancers and to increase the diagnosis sensitivity of radiologists. The CAD system requires a multistage algorithm that includes detection and classification of suspicious lesions.
The region growing technique has been frequently used for segmentation of mammographic masses. The problems with the region growing approach are that it often fails to yield desired results because of difficulties of choosing seed points and an appropriate rule for splitting and merging regions. In order to overcome the problems, this thesis has developed an automatic seed selection method and more robust region growing method for segmentation of mammographic masses, which do not depend on the shape and the absolute intensity values of the mass lesions. As a part of CAD system, the proposed detection method consists of breast region extraction, region partitioning, automatic seed selection, segmentation by region growing, feature extraction, and neural network classification.
The objective performances of the proposed segmentation in the presence of noise were compared with those by other region growing method. Th...