2014 Volume 3 Pages 116-122
We evaluated the performance of our automated computerized scheme for determining the likelihood of malignancy of pulmonary nodules on high-resolution computed tomography (HRCT) and positron emission tomography (PET) scans. Our database contained 36 primary lung cancers and 9 benign nodules. After the nodule location was identified by a radiologist, the boundary surface of the nodule was segmented automatically using a spiral-scanning technique. Objective nodular features were assessed by quantitative analysis of the nodular shape and on gray-level histograms of the interior and exterior regions. The likelihood of malignancy was determined by a support vector machine. The performance of our technique in distinguishing between benign and malignant nodules was evaluated by receiver operating characteristic (ROC) analysis. The area under the ROC curve (AUC) value obtained by using HRCT features alone was 0.87. The integration of PET features into the AUC value resulted in a significant improvement from 0.87 to 0.94 (P< 0.01). The AUC value obtained from simultaneous selection of HRCT and PET features was 0.97. A statistically significant difference (P< 0.01) was observed between the result obtained by simultaneous HRCT and PET feature selection (AUC=0.97) and that by integration of PET features (AUC=0.94). Our automated computerized scheme for determining the likelihood of malignancy may help radiologists to differentiate between benign and malignant pulmonary nodules on HRCT and PET scans.