Autonomous detection of solitary pulmonary nodules on CT images for computer-aided diagnosis

In this paper, algorithms of ROI segmentation, feature selecting and classifying were studied, and a novel scheme has been proposed to detect solitary pulmonary nodules on CT images. ROIs are segmented based on multi-scale morphological filtering method, features of ROI are selected using separability of probability, and ROIs are classified to nodule or non-nodule by improved Mahalanobis distance. Twenty clinical cases were tested in this study, the sensitivity of nodule detection is 94.6%. Experiment results indicated that lung nodule detection using the proposed algorithms is with high sensitivity and low false positive rate, it can provide helpful information for automatic detection of pulmonary nodules in a computer-aided diagnosis(CAD) system.