Shape Analysis 3D to false positive reduction

Abstract

Objective: The present work describes the development of a methodology for the reduction of false positives, so that it helps in the consistent distinction between pulmonary nodules and other structures present in CT examinations. Methods: The proposed method is based on the exploration of characteristics of form related to compactness, spherical disproportion, Feret diameters, skeleton and statistical measures. These characteristics were submitted to a set of classifiers, highlighting the MLP algorithm. The method was evaluated in a set of 24156 regions from 833 examinations of the LIDC-IDRI base. Results: The method was promising, reaching 91% of sensitivity, 92.8% of specificity and 92.3% of accuracy. Conclusion: The results showed that the morphological analysis is satisfactory and presents significant contributions to the CAD systems, directly aiding in a more favorable prognosis to the patient.