SU‐E‐J‐272: Auto‐Segmentation of Regions with Differentiating CT Numbers for Treatment Response Assessment

Abstract

Purpose:It has been reported recently that the change of CT number (CTN) during and after radiation therapy (RT) may be used to assess RT response. The purpose of this work is to develop a tool to automatically segment the regions with differentiating CTN and/or with change of CTN in a series of CTs.Methods:A software tool was developed to identify regions with differentiating CTN using K‐mean Cluster of CT numbers and to automatically delineate these regions using convex hull enclosing method. Pre‐ and Post‐RT CT, PET, or MRI images acquired for sample lung and pancreatic cancer cases were used to test the software tool. K‐mean cluster of CT numbers within the gross tumor volumes (GTVs) delineated based on PET SUV (standard uptake value of fludeoxyglucose) and/or MRI ADC (apparent diffusion coefficient) map was analyzed. The cluster centers with higher value were considered as active tumor volumes (ATV). The convex hull contours enclosing preset clusters were used to delineate these ATVs with color washed displays. The CTN defined ATVs were compared with the SUV‐ or ADC‐defined ATVs.Results:CTN stability of the CT scanner used to acquire the CTs in this work is less than 1.5 Hounsfield Unit (HU) variation annually. K‐mean cluster centers in the GTV have difference of ∼20 HU, much larger than variation due to CTN stability, for the lung cancer cases studied. The dice coefficient between the ATVs delineated based on convex hull enclosure of high CTN centers and the PET defined GTVs based on SUV cutoff value of 2.5 was 90(±5)%.Conclusion:A software tool was developed using K‐mean cluster and convex hull contour to automatically segment high CTN regions which may not be identifiable using a simple threshold method. These CTN regions were reasonably overlapped with the PET or MRI defined GTVs.