Automatic segmentation of glioblastoma for radiation therapy treatment planning

Abstract

During the radiation therapy (RT) process, the treatment is planned and simulated with a treatment planning system (TPS): the organs at risk (OAR) and the tumor target are identified and contoured, and the RT dose, delivered by the planned photon beams, is obtained for optimization of the resulting plan. The contouring work-up of tumor target identifies the Planning Treatment Volume (PTV), i.e. the physical RT treatment volume. PTV of glioblastomas (GB) includes, after expansion, Gross Tumor Volume (GTV, the tumor) and Clinical Target Volume (CTV, tumor plus edema). Usually, GTV contouring is performed manually. In this work, we used GlioCAD, a Computer-Assisted Detection software for automatic contouring of gliomas in MRI/DTI, to delineate GTV. The dataset included the images of 21 patients undergoing RT for GB. For each patient, we co-registered CT-planning images and diagnostic MRI (16 T1-gad, 6 T2 Flair, 13 Flair Fat Sat), which were used for GlioCAD training and validation. CAD outlined the tumor with good accuracy, after ruling out some false positives in post-processing. We identified GTVs, suitable for RT requirements. An evolution of GlioCAD will take into account edema for outlining CTV. The method seems promising. A further automatic system for the delineation of sites at risk in the brain is under development, which may be helpful for standardization of RT-treatment planning.