Lung nodule growth measurement and prediction using auto cluster seed K-means morphological segmentation and shape variance analysis

Abstract

A quantitative model is developed in this work to predict the lung nodules which have the potential to grow in future. An Auto Cluster Seed K-means Morphological segmentation (ACSKMM) algorithm was implemented in this work to segment all the possible lung nodule candidates. An average of around 600 nodule candidates of size >3mm were segmented from each CT scan series of 34 patients. Finally, in total 34 real nodules were remained after eliminating the vessels, non-nodules and calcifications using centroid shift and 3D shape variance analysis. The rate of nodule growth (RNG) was computed on real nodules in terms of 3D-volume change. Out of the 34 real nodules, 3 nodules had RNG value>1, confirming their malignant nature. The nodule growth predictive measure was modelled through compactness, mass deficit, mass excess and isotropic factor.