Malignant nodule detection on lung CT scan images with kernel RX-algorithm

Abstract

In this paper, we present a nonlinear anomaly detector called kernel RX-algorithm and apply it to CT images for malignant nodule detection. Malignant nodule detection is very similar to anomaly detection in military imaging applications where the RX-algorithm has been successfully applied. We modified the original RX-algorithm so that it can be applied to anomaly detection in CT images. Moreover, using kernel trick, we mapped the data to a high dimensional space to obtain a kernelized RX-algorithm that outperforms the original RX-algorithm. The preliminary results of applying the kernel RX-algorithm on annotated public access databases suggests that the proposed method may provide a means for early detection of the malignant nodules.