To explore the diagnostic value of density features in lung tumor screened from low dose computed tomography (LDCT) with thin section. A computer-aided diagnosis (CAD) system was established to assist in defining tumor and density features. Forty-eight surgically confirmed tumors in 38 patients screened by thin-section LDCT were retrospectively enrolled in consecutive manner to examine the performance of this system. The confirmed surgical results included 29 malignant and 19 benign tumors. The pathology of malignancy were adenocarcinoma (AdCa, n=17) and adenocarcinoma in situ (AdIs, n=12). The benignancy included atypical adenomatous hyperplasia (AAH, n=11) and benign non-AAH (n=8). Of density features, tumor Entropy provided the best power to differentiate malignancy from benignancy (p<;0.001), and further to classify the 4-type histopathology (p<;0.001). Feature Entropy has limitation in differentiating AdIs from benign non-AAH, which can be improved using feature of tumor disappearance rate (TDR) and Mean. Entropy and TDR were determined to be best decisive factors in constructing the CAD prediction model, which predicted tumors between malignancy and benignancy with an Az of 0.913. Density features defined using CAD is useful to differentiate malignancy from benignancy of lung tumors screened using thin-section multi-detector LDCT, and further to predict histopathology.
Relation:
the sixth International Conference on Genetic and Evolutionary Computing
