Abstract
To improve the performance of the computer-aided systems for breast cancer diagnosis, the ensemble classifier is proposed for classifying the histological structures in the breast cancer microscopic images into three region types: positive cancer cells, negative cancer cells and non-cancer cell (stromal cells and lymphocyte cells) image. The bagging and boosting ensemble techniques are used with the decision tree (DT) learner. They are also compared with the single classifier, DT. The feature used as an input of classifiers is the fractal dimension (FD) based 12 color channels. It is computed from the image datasets, which are manually prepared in small cropped image with 3 window sizes including 128×128 pixels, 192×192 pixels and 256×256 pixels. The results show that the boosting ensemble classifier gives the best accuracy about 80% from window size of 256, although it is the lowest when using the single DT as classifier. The results indicated that the ensemble method is capable of improving the accuracy in the classification compared to the single classifier. The classification model using FD and the ensemble classifier would be applied to develop the computer- aided systems for breast cancer diagnosis in the future.