Purpose/Objective(s)An online treatment process which integrates the simulation, planning and delivery steps in one single session of ≤ 30 min on a treatment unit capable of cone-beam CT imaging (CBCT) is under development in our institution for patients with spinal bone metastases. This method of treatment, if performed in a reasonable time, has the potential to reduce patient discomfort, increase department efficiency, and consequently improve patient access to palliative radiotherapy. The objective of this work is to streamline and provide for a consistent online planning process on cone-beam CT datasets through semi-automatic localization and segmentation of bony target.Materials/MethodsThe semi-automatic vertebrae segmentation consisted of two steps: the radiographic localization (2D) and the volumetric segmentation. The radiographic localization of vertebrae was performed on coronal and sagittal digital reconstructed radiographs (DRR). These were produced by projecting the CBCT images orthogonally to the manually defined centerline of a region of interest (ROI), which encompassed the spine and followed its curvature. After detection, the vertebrae were numbered starting from the skull or the sacrum. For thoracic targets, a low (0.5 cGy) and a high dose (2 cGy) scan acquired with a table shift could be joined together to increase the CBCT field-of-view and include one of the spine extremities. The volumetric segmentation of the bony target localized on the DRRs was performed using a model-based algorithm on the 3D dataset. The validation of the semi-automatic vertebrae localization and segmentation was done on patient data obtained from a protocol approved by research ethics board. The success of the radiographic localization was measured as the number of vertebrae localized within the area of their vertebral body. The accuracy of the volumetric segmentation was assessed by compiling the distances between each point on the surface of the auto-segmentation and it’s closest counterpart on the manual contours. The duration of both tasks was also measured.ResultsFor 18 patients and a total of 129 vertebrae, 95.3% of the vertebrae were successfully localized on DRR (≤ 1 min, low and high dose scans). The volumetric segmentation was validated on a subset of 10 patients; most of them presenting deteriorating bone disease. For 26 vertebrae (cervical to sacral spine), the average linear discrepancy between auto and manual segmentations was 1.3 mm ± 1.4 mm (RMS of standard deviations). The volumetric segmentation included the placement of the model at the level of the target (translation and rotation only) and was performed on average within 1.6 ± 0.4 min (SD). The use of manual model rescaling and deformation prior to the auto-segmentation increased the length of the process by 1.9 min and did not improve significantly its accuracy.ConclusionsAutomatic localization and segmentation of vertebrae for patients with spinal bone metastases was shown to be robust and effective on clinical CBCT images. These tools can reduce the duration of the online planning and treatment process and potentially improve the consistency of target definition in palliative radiotherapy. Challenges not yet addressed include the management of potential soft tissue extension around the bony target. Use of prior diagnostic information and development of margin expansion tools are under investigation. Purpose/Objective(s)An online treatment process which integrates the simulation, planning and delivery steps in one single session of ≤ 30 min on a treatment unit capable of cone-beam CT imaging (CBCT) is under development in our institution for patients with spinal bone metastases. This method of treatment, if performed in a reasonable time, has the potential to reduce patient discomfort, increase department efficiency, and consequently improve patient access to palliative radiotherapy. The objective of this work is to streamline and provide for a consistent online planning process on cone-beam CT datasets through semi-automatic localization and segmentation of bony target. An online treatment process which integrates the simulation, planning and delivery steps in one single session of ≤ 30 min on a treatment unit capable of cone-beam CT imaging (CBCT) is under development in our institution for patients with spinal bone metastases. This method of treatment, if performed in a reasonable time, has the potential to reduce patient discomfort, increase department efficiency, and consequently improve patient access to palliative radiotherapy. The objective of this work is to streamline and provide for a consistent online planning process on cone-beam CT datasets through semi-automatic localization and segmentation of bony target. Materials/MethodsThe semi-automatic vertebrae segmentation consisted of two steps: the radiographic localization (2D) and the volumetric segmentation. The radiographic localization of vertebrae was performed on coronal and sagittal digital reconstructed radiographs (DRR). These were produced by projecting the CBCT images orthogonally to the manually defined centerline of a region of interest (ROI), which encompassed the spine and followed its curvature. After detection, the vertebrae were numbered starting from the skull or the sacrum. For thoracic targets, a low (0.5 cGy) and a high dose (2 cGy) scan acquired with a table shift could be joined together to increase the CBCT field-of-view and include one of the spine extremities. The volumetric segmentation of the bony target localized on the DRRs was performed using a model-based algorithm on the 3D dataset. The validation of the semi-automatic vertebrae localization and segmentation was done on patient data obtained from a protocol approved by research ethics board. The success of the radiographic localization was measured as the number of vertebrae localized within the area of their vertebral body. The accuracy of the volumetric segmentation was assessed by compiling the distances between each point on the surface of the auto-segmentation and it’s closest counterpart on the manual contours. The duration of both tasks was also measured. The semi-automatic vertebrae segmentation consisted of two steps: the radiographic localization (2D) and the volumetric segmentation. The radiographic localization of vertebrae was performed on coronal and sagittal digital reconstructed radiographs (DRR). These were produced by projecting the CBCT images orthogonally to the manually defined centerline of a region of interest (ROI), which encompassed the spine and followed its curvature. After detection, the vertebrae were numbered starting from the skull or the sacrum. For thoracic targets, a low (0.5 cGy) and a high dose (2 cGy) scan acquired with a table shift could be joined together to increase the CBCT field-of-view and include one of the spine extremities. The volumetric segmentation of the bony target localized on the DRRs was performed using a model-based algorithm on the 3D dataset. The validation of the semi-automatic vertebrae localization and segmentation was done on patient data obtained from a protocol approved by research ethics board. The success of the radiographic localization was measured as the number of vertebrae localized within the area of their vertebral body. The accuracy of the volumetric segmentation was assessed by compiling the distances between each point on the surface of the auto-segmentation and it’s closest counterpart on the manual contours. The duration of both tasks was also measured. ResultsFor 18 patients and a total of 129 vertebrae, 95.3% of the vertebrae were successfully localized on DRR (≤ 1 min, low and high dose scans). The volumetric segmentation was validated on a subset of 10 patients; most of them presenting deteriorating bone disease. For 26 vertebrae (cervical to sacral spine), the average linear discrepancy between auto and manual segmentations was 1.3 mm ± 1.4 mm (RMS of standard deviations). The volumetric segmentation included the placement of the model at the level of the target (translation and rotation only) and was performed on average within 1.6 ± 0.4 min (SD). The use of manual model rescaling and deformation prior to the auto-segmentation increased the length of the process by 1.9 min and did not improve significantly its accuracy. For 18 patients and a total of 129 vertebrae, 95.3% of the vertebrae were successfully localized on DRR (≤ 1 min, low and high dose scans). The volumetric segmentation was validated on a subset of 10 patients; most of them presenting deteriorating bone disease. For 26 vertebrae (cervical to sacral spine), the average linear discrepancy between auto and manual segmentations was 1.3 mm ± 1.4 mm (RMS of standard deviations). The volumetric segmentation included the placement of the model at the level of the target (translation and rotation only) and was performed on average within 1.6 ± 0.4 min (SD). The use of manual model rescaling and deformation prior to the auto-segmentation increased the length of the process by 1.9 min and did not improve significantly its accuracy. ConclusionsAutomatic localization and segmentation of vertebrae for patients with spinal bone metastases was shown to be robust and effective on clinical CBCT images. These tools can reduce the duration of the online planning and treatment process and potentially improve the consistency of target definition in palliative radiotherapy. Challenges not yet addressed include the management of potential soft tissue extension around the bony target. Use of prior diagnostic information and development of margin expansion tools are under investigation. Automatic localization and segmentation of vertebrae for patients with spinal bone metastases was shown to be robust and effective on clinical CBCT images. These tools can reduce the duration of the online planning and treatment process and potentially improve the consistency of target definition in palliative radiotherapy. Challenges not yet addressed include the management of potential soft tissue extension around the bony target. Use of prior diagnostic information and development of margin expansion tools are under investigation.