Hybrid intelligent modeling and prediction of texture segmented lesion from 4DCT scans of thorax

In this study, modeling and prediction from four dimensional computed tomography images of texture delineated sub-lesion region by using hybrid intelligent algorithm (adaptive neural fuzzy inference system) is presented. Texture segmented sub-lesion region was segmented by fuzzy C means clustering and deformation maps between segmented regions are computed by using an expectation minimization approach. Both rigid (global) and non-rigid (local) registration was performed on the data. The data consisted of 4 phases of respiratory cycle totalling 36 images containing sub-lesion regions. Parameters extracted from the maps were fed to the hybrid intelligent algorithm for training and validation. The root mean square error for modeling and prediction was 10^-7 for rigid parameter modeling and 46.06 for non-rigid modeling, respectively. The artificial sequence of sub-lesion regions was warped by using predicted parameters from the hybrid intelligent algorithm. The artificially generated warped images and true segmented images were then compared. The registration error was determined by the correlation coefficient and was found to be 0.603.