Interactive Labeling System for Lung Nodules with CT Images

Abstract

Survival rate of lung cancer is up to 70%~90% in 5 years, if malignant lung nodules are found early. The lung nodules are malignant or benign based on size changes in a short time such as two or three months. Therefore, the volumes of lung nodules need be known. Until now, radiologists use some software to measure the volumes of lung nodules such as ImageJ, Label me and so on. These software are difficult to precisely draw the contour of a lung nodule as radiologists' thinking. Especially to the solid part of Ground-Glass Opacity nodule, the software are difficult to measure its volume. In this study, an interactive labeling system is developed for lung nodules with CT images. The proposed interactive labeling system provides three ways to draw the contour of a lung nodule: $\bigcirc\!\!\!\!\!1$. To draw the contour of a lung nodule manually. $\bigcirc\!\!\!\!\!2$. To draw the contour of a lung nodule with a semi-automatic way. $\bigcirc\!\!\!\!\!3$. To draw the contour of a lung nodule with a semiautomatic way and then modify the contour of the lung nodule manually. Moreover, radiologists can label the lung nodule based on its malignancy or benign and give some comments to the lung nodule by using the proposed labeling system. The annotation of the lung nodule by radiologists can be saved as JavaScript Object Notation (JSON) format. Therefore, there are two meanings on ‘label’ of interactive labeling system. One is to draw the contour of a lung nodule, and the other is to give a label on malignancy or benign of a lung nodule. Practicality and effectiveness of the interactive labeling system are verified by the questionnaire.