ROC-like optimization by sample ranking: Application to CT colonography

In this study, we propose a new binary classification algorithm which optimizes the area under the receiver operating characteristic curve (AUC) based on a ranking of training samples. We first solved the traditional AUC maximization problem using semi-definite programming. We then introduced auxiliary variables to rank training samples. In this way, we can select the most representative samples to avoid over-training to noisy data. We applied our proposed classifier to a CT colonography dataset containing 50 patients. Preliminary experimental results indicate that our proposed method can achieve higher classification performance than support vector machines.