Combining multiple 2ν-SVM classifiers for tissue segmentation

In image classification problems, especially those involving tumor or precancerous lesion, we are usually faced with the situation in which the cost of mistakenly classifying samples in one class is much higher than that of the opposite mistake in the other class. Therefore it is essential to include cost information about classes in our classification methods. This paper applies a cost-sensitive 2ν-SVM classification scheme to cervical cancer images to separate diseased regions from healthy tissue. Using this method, we are able to specify a higher weight to the class that is deemed more important. To the best of our knowledge, cost-sensitive SVM based medical image classification has not been done before. We specifically target segmenting disease regions in digitized uterine cervix images in a NCI/NLM archive of 60,000 images. Our second contribution is the introduction of a multiple classifier scheme instead of the traditional single classifier model. Using the multiple classi-fier scheme improves significantly classification accuracy as demonstrated by our experiments.