Automated delineation of brain structures in patients undergoing radiotherapy for primary brain tumors: From atlas to dose–volume histograms

Purpose: To implement and evaluate a magnetic resonance imaging atlas-based automated segmentation (MRI-ABAS) procedure for cortical and sub-cortical grey matter areas definition, suitable for dose-distribution analyses in brain tumor patients undergoing radiotherapy (RT). Patients and methods: 3T-MRI scans performed before RT in ten brain tumor patients were used. The MRI-ABAS procedure consists of grey matter classification and atlas-based regions of interest definition. The Simultaneous Truth and Performance Level Estimation (STAPLE) algorithm was applied to structures manually delineated by four experts to generate the standard reference. Performance was assessed comparing multiple geometrical metrics (including Dice Similarity Coefficient – DSC). Dosimetric parameters from dose–volume-histograms were also generated and compared. Results: Compared with manual delineation, MRI-ABAS showed excellent reproducibility [median DSC_ABAS = 1 (95% CI, 0.97–1.0) vs. DSC_MANUAL = 0.90 (0.73–0.98)], acceptable accuracy [DSC_ABAS = 0.81 (0.68–0.94) vs. DSC_MANUAL = 0.90 (0.76–0.98)], and an overall 90% reduction in delineation time. Dosimetric parameters obtained using MRI-ABAS were comparable with those obtained by manual contouring. Conclusions: The speed, reproducibility, and robustness of the process make MRI-ABAS a valuable tool for investigating radiation dose–volume effects in non-target brain structures providing additional standardized data without additional time-consuming procedures