Combining one-class classifiers for imbalanced classification of breast thermogram features

Abstract

Thermography provides an interesting modality for diagnosing breast cancer as it is a non-contact, non-invasive and passive technique that is able to detect small tumors, which in turn can lead to earlier diagnosis. We perform computer-aided diagnosis of breast thermograms based on image features describing bilateral differences in regions of interest and a pattern classification approach that learns from previous examples. As is often the case in medical diagnosis, such training sets are imbalanced as typically (many) more benign cases get recorded compared to malignant cases. In this paper, we address this problem and perform classification using an ensemble of one-class classifiers. One-class classification uses samples from a single distribution to derive a decision boundary, and employing this method on the minority class can significantly boost its recognition rate and hence the sensitivity of our approach. We combine several one-class classifiers using a random subspace approach and a diversity measure to select members of the committee. We show that our proposed technique works well and leads to significantly improved performance compared to a single one-class predictor as well as compared to state-of-the-art classifier ensembles for imbalanced data.