Breast Thermogram Analysis Using Classifier Ensembles and Image Symmetry Features

Breast cancer is the most commonly diagnosed form of cancer in women. Thermography, which uses cameras with sensitivities in the thermal infrared, has been shown to provide an interesting alternative to the standard modality of mammography for detecting breast cancer as it is able to detect smaller tumors and hence can lead to earlier diagnosis. In this paper, we present an approach to breast thermogram analysis that extracts features describing bilateral symmetries from an image and then utilizes a classifier ensemble for decision making. Importantly, our classification approach addresses the problem of imbalanced class distribution that is common in medical data analysis. We do this by constructing feature subspaces from balanced data subsets and train different classifiers on different subspaces. To combine the individual classifiers, we investigate two different strategies. The first dynamically assigns classifier weights based on an evolutionary algorithm, while the second uses a neural network for classifier fusion. Both approaches are shown to work well and to lead to significantly improved performance compared to canonical classification systems.