Computer classification of lung tumors from chest CT images according to the types of tissue using 3D extended Voronoi diagram

Abstract

A method to quantify the shrinkage of lung lobe using chest X-ray CT images is proposed in this paper. The shrinkage of lung lobe is caused by the convergence concerning the interstitium. It is known that the existence of the convergence gives us useful information about the malignancy of tumor. If the convergence occurs, vessels are distributed densely in the neighborhood of tumor. In this paper, the volume of the Voronoi diagrams using the vessel regions are calculated. We propose two features using the difference of the volumes of the Voronoi regions near the tumor region and apart from the tumor region. Features are calculated using seventy-eight cases. First, as the result of the significance test using twenty-nine cases for which the degree of the convergence was evaluated by medical doctors, it is shown that both the two features tend to have high values when the convergence exists. Next, using other twenty-five cases, the proposed feature cans classify the cases into types of the tissue which tend to cause the convergence or not. Finally, The result of benign/malignant discrimination using seventy-eight cases was performed. False positive rate was 0.26(7/27) and false negative rate was 0.14(7/51).