- Email: [email protected]
Computer rendering of radiation field light anatomic position for the quality assurance (QA) of linear accelerator based radiosurgery (SRS)
I. J. Radiation O n c o l o g y • B i o l o g y • Physics
364
V o l u m e 42, N u m b e r 1 S u p p l e m e n t , 1998
2273 VOLUME RENDERING S E G M E N T A T I O N A L G O R I T H M (VRSA) FOR SYSTEMATIC COMPARISON OF V I S U A L I Z A T I O N T E C H N I Q U E S FOR CEREBRAL ARTERIOVENOUS MALFORMATIONS Jani, Ashesh B., M.D., M.S.E.E.; Pelizzari, Charles A., Ph.D.; Chen, George T.Y., Ph.D.; Roeske, John C., Ph.D.; Hamilton, Russell J., Ph.D.; *MacDonald, Loch, M.D., Ph.D.; **Bova, Frank, Ph.D.; and Sweeney, Patrick, M.D. Department of Radiation and Cellular Oncology and *Department of Neurosurgery, University of Chicago, Chicago, IL, USA; ** Department of Radiation Oncology, University of Florida, Gainesville, FL, USA Objective: To describe Volume Rendering Segmentation Algorithm (VRSA), a new technique for volumetric analysis of volume-rendered images, and to report the results obtained when applying VRSA to contrast-enhanced cerebral arteriovenoas malformations (AVM's). Volume rendering is a visualization technique that inherently preserves anatomic information in spatial context. Although volume-rendered images are rich in anatomic detail, comparison of these images to conventional axially-defined images has not yet been undertaken systematically and quantitatively due to the lack of volumetric analysis tools. A cerebral AVM, which is a high-contrast structure for which precise volume definition is critical to avoid treatment toxicity when performing stereotactic radiosurgery, is an ideal initial structure for testing and calibration of VRSA. Methods: Six cases of cerebral AVM's (representing a wide range of AVM size, shape, and location) for which biplanar angiograms and power-inject contrast-enhanced CT data exist were analyzed for this investigation. For each case, a neurosurgeon and a radiation oncologist defined the AVM nidus on the biplanar angiograms, on the axial CT slices, and on volume-rendered images of the AVM captured in each of the six principle viewing directions (antero-posterior, postero-anterior, right lateral, left lateral, supero-inferior, and infero-superior). VRSA is a three-step algorithm that involves (a) construction of six principle partial surfaces from the nidus definition on each of the volume-rendered images, (b) processing and smoothing of each of the partial surfaces, and (c) merging these six modified partial surfaces together to create a volume. VRSA was tested on spherical CT phantoms prior to its application to the patient data. The AVM volumes obtained using VRSA were compared to those obtained using axial CT and to those obtained using biplanar angiograms that were correlated to CT. Results: Qualitatively, the volume-rendered images preserved more anatomic detail of the AVM nidus in spatial context with the feeding arteries, draining veins, and brain parenchyma than did the axial CT images. Also, the AVM nidus definition on the volume-rendered images was, by virtue of requiring the definition of only six regions of interest (one for each of the principle viewing directions), considerably less time consuming for the clinician than definition on axial CT, particularly for the larger AVM' s. Furthermore, the volume-rendered images avoided the planar projection limitations of CT-correlated angiography. Quantitatively, on the phantom data sets, the volumes obtained using VRSA compared to within 5% of the axially-defined volumes. On the actual AVM cases, the volumes obtained using VRSA compared to within 8% of the axially-defined volumes, and to within 5% of the angiographicallydefined volumes, for all the analyzed cases. The volumes obtained using VRSA were generally smaller than the angiographically-defined volumes (which is likely due to the volume overestimation problem intrinsic to angiogram-CT correlation), and generally larger than the axially-defined volumes (which is likely due to improved visualization of the regional anatomy on the volume-rendered images).
Conclusions: The qualitative results support the use of volume rendering, which preserves more visual anatomic information in spatial context, over both angiography and axial CT for AVM nidus definition. The AVM volumes obtained using VRSA compare very favorably with and are considerably less timeconsuming to produce than those obtained using the conventional visualization techniques. Although the specific application for this investigation was to the cerebral AVM, VRSA has general applicability to volumetric and dosimetric analysis at many disease sites for which precise target volume and critical structure definition is necessary. VRSA represents an essential development for the clinical use of volume rendering for treatment planning.
2274 C O M P U T E R RENDERING OF R A D I A T I O N FIELD L I G H T A N A T O M I C P O S I T I O N FOR T H E Q U A L I T Y ASSURANCE (QA) OF L I N E A R A C C E L E R A T O R BASED R A D I O S U R G E R Y (SRS) Engler, Mark J., Ph.D., Pagnini, Paul G., M.D., DiPetrillo, Thomas, M.D., Wazer, David, M.D., Supran, Stacy M.S.,* and Tsai, Jen-San, Ph.D. Department of Radiation Oncology and *Division of Biostatistics, New England Medical Center & Tufts Univ. School of Medicine, Boston, MA
Objectives: To apply the rendering of planned radiosurgical field light anatomic position on patient surface anatomy as a quality assurance (QA) indicator for linear accelerator (linac) based SRS, in lieu of port films. Materials & Methods: A 3-dimensional linac-based SRS treatment planning and delivery system has been applied to over 140 patients over a period of eight years. Tertiary SRS collimators have been applied in conjunction with two megavoltage linacs. Port films have not been obtained because of excessive time and risk required to remove SRS collimators near the patient and obtain a large field exposure including an image of the SRS field. Consequently a substitute QA indicator for treatment localization with the patient in treatment position was developed. The treatment planning system is applied to render views of 3-dimensional patient surface anatomy and the SRS headframe based on computerized tomography (CT) voxels typically 3 x 0.7 x 0.7 m m in vertical (V), anterior-posterior (AP), and lateral (L) dimensions, respectively, and from the viewpoints of~townward and horizontally pointing linac gantry angles (00 and 01). The planning system is then manipulated to render the intended outline of the radiation field light on the patient's surface anatomy and/or posts and pins of the headframe. Distances (d) from headframe reference points to edges and centers of beam spots are measured on the computer renderings, and subtracted from measurements made with rulers on patient cranial surfaces with the patient in treatment position, to create a metric Adl, where i denotes the AP, L, or V dimension. At least two values of Ad in two different dimensions, one at 00 and a second at 0,, are measured per patient. The values of Ad are written on the rendering as documentation of agreement between the treatment planning and treatment room beam geometries. An absolute value lad I _< 2 m m was set as a criterion for triggering a double-check of image registration and other aspects of treatment planning. Over 300 values of Ad from clinical treatments were analyzed to search for statistically significant effects relative to: the magnitude IAdl, the sign of Ad, the older versus newer linac, and the anatomic direction (AP, Lat or V). Results: During the initial years of the SRS program, three IAdl >> 2 mm revealed treatment planning image registration errors, thus averting potential SRS fatalities. New plans obtained with corrected image registration yielded IAdl -< 2 mm, allowing the patients to be treated. Statistically significant differences of IAdl relative to the magnitude tdl were observed, and attributed to a larger experimental error in measuring these values with rulers extending over larger curvature of the patients' cranial surfaces. The sign of Ad was fotmd to be random. Statistically significant increased IAdl of the newer relative to the older linac were attributed to the greater weight and isocenter motion of the newer linac's gantry. Statistically significant differences oflAdvl relative to IAdAPIand IAdLIwere also observed. This observation may be attributed to effects on the computer rendering accuracy that may arise from the typically large CT slice width of 3 mm in the V dimension relative to submillimeter pixel dimensions in the AP and L dimensions. Conclusion: Application of computer renderings of linac field outlines on patient surface anatomy has functioned successfully as a QA indicator for linac based SRS.
364
V o l u m e 42, N u m b e r 1 S u p p l e m e n t , 1998
2273 VOLUME RENDERING S E G M E N T A T I O N A L G O R I T H M (VRSA) FOR SYSTEMATIC COMPARISON OF V I S U A L I Z A T I O N T E C H N I Q U E S FOR CEREBRAL ARTERIOVENOUS MALFORMATIONS Jani, Ashesh B., M.D., M.S.E.E.; Pelizzari, Charles A., Ph.D.; Chen, George T.Y., Ph.D.; Roeske, John C., Ph.D.; Hamilton, Russell J., Ph.D.; *MacDonald, Loch, M.D., Ph.D.; **Bova, Frank, Ph.D.; and Sweeney, Patrick, M.D. Department of Radiation and Cellular Oncology and *Department of Neurosurgery, University of Chicago, Chicago, IL, USA; ** Department of Radiation Oncology, University of Florida, Gainesville, FL, USA Objective: To describe Volume Rendering Segmentation Algorithm (VRSA), a new technique for volumetric analysis of volume-rendered images, and to report the results obtained when applying VRSA to contrast-enhanced cerebral arteriovenoas malformations (AVM's). Volume rendering is a visualization technique that inherently preserves anatomic information in spatial context. Although volume-rendered images are rich in anatomic detail, comparison of these images to conventional axially-defined images has not yet been undertaken systematically and quantitatively due to the lack of volumetric analysis tools. A cerebral AVM, which is a high-contrast structure for which precise volume definition is critical to avoid treatment toxicity when performing stereotactic radiosurgery, is an ideal initial structure for testing and calibration of VRSA. Methods: Six cases of cerebral AVM's (representing a wide range of AVM size, shape, and location) for which biplanar angiograms and power-inject contrast-enhanced CT data exist were analyzed for this investigation. For each case, a neurosurgeon and a radiation oncologist defined the AVM nidus on the biplanar angiograms, on the axial CT slices, and on volume-rendered images of the AVM captured in each of the six principle viewing directions (antero-posterior, postero-anterior, right lateral, left lateral, supero-inferior, and infero-superior). VRSA is a three-step algorithm that involves (a) construction of six principle partial surfaces from the nidus definition on each of the volume-rendered images, (b) processing and smoothing of each of the partial surfaces, and (c) merging these six modified partial surfaces together to create a volume. VRSA was tested on spherical CT phantoms prior to its application to the patient data. The AVM volumes obtained using VRSA were compared to those obtained using axial CT and to those obtained using biplanar angiograms that were correlated to CT. Results: Qualitatively, the volume-rendered images preserved more anatomic detail of the AVM nidus in spatial context with the feeding arteries, draining veins, and brain parenchyma than did the axial CT images. Also, the AVM nidus definition on the volume-rendered images was, by virtue of requiring the definition of only six regions of interest (one for each of the principle viewing directions), considerably less time consuming for the clinician than definition on axial CT, particularly for the larger AVM' s. Furthermore, the volume-rendered images avoided the planar projection limitations of CT-correlated angiography. Quantitatively, on the phantom data sets, the volumes obtained using VRSA compared to within 5% of the axially-defined volumes. On the actual AVM cases, the volumes obtained using VRSA compared to within 8% of the axially-defined volumes, and to within 5% of the angiographicallydefined volumes, for all the analyzed cases. The volumes obtained using VRSA were generally smaller than the angiographically-defined volumes (which is likely due to the volume overestimation problem intrinsic to angiogram-CT correlation), and generally larger than the axially-defined volumes (which is likely due to improved visualization of the regional anatomy on the volume-rendered images).
Conclusions: The qualitative results support the use of volume rendering, which preserves more visual anatomic information in spatial context, over both angiography and axial CT for AVM nidus definition. The AVM volumes obtained using VRSA compare very favorably with and are considerably less timeconsuming to produce than those obtained using the conventional visualization techniques. Although the specific application for this investigation was to the cerebral AVM, VRSA has general applicability to volumetric and dosimetric analysis at many disease sites for which precise target volume and critical structure definition is necessary. VRSA represents an essential development for the clinical use of volume rendering for treatment planning.
2274 C O M P U T E R RENDERING OF R A D I A T I O N FIELD L I G H T A N A T O M I C P O S I T I O N FOR T H E Q U A L I T Y ASSURANCE (QA) OF L I N E A R A C C E L E R A T O R BASED R A D I O S U R G E R Y (SRS) Engler, Mark J., Ph.D., Pagnini, Paul G., M.D., DiPetrillo, Thomas, M.D., Wazer, David, M.D., Supran, Stacy M.S.,* and Tsai, Jen-San, Ph.D. Department of Radiation Oncology and *Division of Biostatistics, New England Medical Center & Tufts Univ. School of Medicine, Boston, MA
Objectives: To apply the rendering of planned radiosurgical field light anatomic position on patient surface anatomy as a quality assurance (QA) indicator for linear accelerator (linac) based SRS, in lieu of port films. Materials & Methods: A 3-dimensional linac-based SRS treatment planning and delivery system has been applied to over 140 patients over a period of eight years. Tertiary SRS collimators have been applied in conjunction with two megavoltage linacs. Port films have not been obtained because of excessive time and risk required to remove SRS collimators near the patient and obtain a large field exposure including an image of the SRS field. Consequently a substitute QA indicator for treatment localization with the patient in treatment position was developed. The treatment planning system is applied to render views of 3-dimensional patient surface anatomy and the SRS headframe based on computerized tomography (CT) voxels typically 3 x 0.7 x 0.7 m m in vertical (V), anterior-posterior (AP), and lateral (L) dimensions, respectively, and from the viewpoints of~townward and horizontally pointing linac gantry angles (00 and 01). The planning system is then manipulated to render the intended outline of the radiation field light on the patient's surface anatomy and/or posts and pins of the headframe. Distances (d) from headframe reference points to edges and centers of beam spots are measured on the computer renderings, and subtracted from measurements made with rulers on patient cranial surfaces with the patient in treatment position, to create a metric Adl, where i denotes the AP, L, or V dimension. At least two values of Ad in two different dimensions, one at 00 and a second at 0,, are measured per patient. The values of Ad are written on the rendering as documentation of agreement between the treatment planning and treatment room beam geometries. An absolute value lad I _< 2 m m was set as a criterion for triggering a double-check of image registration and other aspects of treatment planning. Over 300 values of Ad from clinical treatments were analyzed to search for statistically significant effects relative to: the magnitude IAdl, the sign of Ad, the older versus newer linac, and the anatomic direction (AP, Lat or V). Results: During the initial years of the SRS program, three IAdl >> 2 mm revealed treatment planning image registration errors, thus averting potential SRS fatalities. New plans obtained with corrected image registration yielded IAdl -< 2 mm, allowing the patients to be treated. Statistically significant differences of IAdl relative to the magnitude tdl were observed, and attributed to a larger experimental error in measuring these values with rulers extending over larger curvature of the patients' cranial surfaces. The sign of Ad was fotmd to be random. Statistically significant increased IAdl of the newer relative to the older linac were attributed to the greater weight and isocenter motion of the newer linac's gantry. Statistically significant differences oflAdvl relative to IAdAPIand IAdLIwere also observed. This observation may be attributed to effects on the computer rendering accuracy that may arise from the typically large CT slice width of 3 mm in the V dimension relative to submillimeter pixel dimensions in the AP and L dimensions. Conclusion: Application of computer renderings of linac field outlines on patient surface anatomy has functioned successfully as a QA indicator for linac based SRS.