Cervical cancer detection using SVM based feature screening

We present a novel feature screening algorithm by deriving relevance measures from the decision boundary of Support Vector Machines. It alleviates the "independence" assumption of traditional screening methods, e.g. those based on Information Gain and Augmented Variance Ratio, without sacrificing computational efficiency. We applied the proposed method to a bottom-up approach for automatic cervical cancer detection in multispectral microscopic thin PAP smear images. An initial set of around 4,000 multispectral texture features is effectively reduced to a computationally manageable size. The experimental results show significant improvements in pixel-level classification accuracy compared to traditional screening methods.