An automated multimodal spectral cluster based segmentation for tumor and lesion detection in PET images

Abstract

The acquisition of Positron Emission Tomography (PET) images for tumor and lesion detection has emerged as one of the most powerful tools for medical image analysis in recent years. In this paper, a novel technique to obtain multimodality aspect of tumor and lesion detection in PET images through Automated Multimodal Spectral Cluster based Segmentation (AMSCS) is proposed, aiming at improving the tumor detection accuracy. The Spectral Contours with Constrained Threshold (SCCT) technique in AMSCS is carried out to various spectral features of the PET image without any deformation, improving the true positive rate. The SCCT technique utilize user defined seed point in the region of interest in PET images and generate spectral contours (i.e.,) shape, size, location and intensity. A Multi-Spectral Contour Cluster (MSCC) mechanism is introduced that organizes the spectral contour features of shape, size, location and intensity into multi-spectral clusters for quicker segmentation of PET Image regions of interest. Experimental analysis is conducted using Primary Tumor Data Set from UCI repository PET Images on parametric such as, Multi-spectral cluster size, ROI segmentation time, tumor and lesion detection time, tumor detection accuracy.