- Email: [email protected]
248 oral 4DCT – HIGH QUALITY IMAGING AND CONTRAST ENHANCEMENT FOR 3D RADIOTHERAPY TREATMENT PLANNING
S 96
P ROFFERED PAPER
T UESDAY, M AY 10, 2011
248 oral 4DCT – HIGH QUALITY IMAGING AND CONTRAST ENHANCEMENT FOR 3D RADIOTHERAPY TREATMENT PLANNING C. Schneider1 , J. Nijkamp1 , S. Rit1 , M. van Herk1 , J. J. Sonke1 1 T HE N ETHERLANDS C ANCER I NSTITUTE - A NTONI VAN L EEUWENHOEK H OSPITAL, Amsterdam, Netherlands
Conclusions: The visualization of actual dose region delivered in a treatment has been available by means of VMAT treatment. The VMAT-CT superimposed on in-VMAT kV-CBCT could be a visual verification of the actual treatment in the high-precision treatment era. 247 oral THE USE OF A GLOBAL RESPIRATORY MOTION MODEL FOR REAL TIME 4D RADIOTHERAPY APPLICATIONS H. Fayad1 , T. Pan2 , D. Visvikis1 1
INSERM U650 L ATIM, Brest, France MD A NDERSON C ANCER C ENTER, Department of physical imaging, Houston, USA 2
Purpose: Respiratory motion modeling is a key aspect for improving radiation therapy, with the objective of delivering less dose to the normal tissues and higher dose to the tumor during the breathing motion cycle. The main idea of this work is the use a generic respiratory motion model to predict in real-time the respiratory motion of the tumor and the adjacent anatomy. Materials: The generic respiratory motion model used in this work is based on principal component analysis. Only two static CT images (end-inspiration, end-expiration) in combination with respiratory synchronized patient’s body surface are needed to adapt the model on a given patient anatomy. This generic model relates the patient’s surface acquired by an external camera and the internal motion calculated using a non rigid registration algorithm. The resulting patient adapted model is subsequently used to generate in realtime the motion of the tumor and the adjacent anatomy. In this study 6 patients were used to create the model and 4 other patients were used in the validation step. The intrinsic model efficiency was tested by comparing model generated 4D-CT volumes and their corresponding acquired 4D-CT volumes. The model based motion information is subsequently used to improve planning and delivery of 4D-IMRT by accounting for 3D tumor (Suh et al 2009) as well as normal tissue motion. The whole schema of the procedure is shown in figure 1.
Results: Good correlation (r=0.93±0.4) and expert validation (1.9mm±0.6mm: based on expert anatomical landmarks identification) was found between the model generated series and the acquired 4D-CT volumes. Preliminary results indicate that including the normal tissue motion provided by our motion model during the delivery of the 4D-IMRT is better in terms of dose delivery optimization than simply accounting for tumor motion. Conclusions: The use of a global respiratory motion model, which may be adapted on a specific patient, for radiotherapy applications avoids unnecessary 4D-CT acquisitions. Accounting for normal tissue motion helps to optimize 4D dosimetry compared to considering the tumor motion only.
Purpose: Planning of radiotherapy for moving tumours in lung or abdomen can be based on respiration-correlated CT-imaging of the tumour in a time weighted averaged (TWA) position. Locating the closest bin in the respiration period corresponding to the TWA position (MidV-CT) has drawbacks : irregular breathing causes artifacts, while signal to noise ratios and contrast enhancement (CE) are poorer than in standard 3D-CT. These drawbacks can be overcome by deforming all 4D image data available to a TWA-3D scan to a Mid-Position (MidP)-CT, as proposed by Wolthaus et al [1]. The purpose of this study is to evaluate the effect of this method on image quality. Materials: For evaluation of the new technique, 10 patients were selected with a clearly visible tumour in the periphery of the lung. For all patients, a 4D-CT and an expiration breath-hold (BH) CT were acquired. For 5 of these patients, contrast was given during BH-CT. The 4D-CT-scans were reconstructed in 10 bins over the respiration cycle.The resulting 10 image sets were deformable registered, applying an iterative multi-scale phase-based optical flow estimation procedure. For each patient 2 different registrations were performed. First, the image set at maximum expiration was used as reference scan to which all other image sets were registered. Next, the average Displacement Vector Field (DVF) over all image sets was subtracted from each DVF yielding all deformations with respect to the TWA position. These DVF were applied to deform all image sets to the TWA position. Finally, the MidP-CT was calculated as median grey-value for each pixel over all 10 image sets. In the second procedure, an additional DVF from the BH-CT to the TWA position was calculated. With the latter, a MidP-BH-CT was calculated by deforming the BH-CT to the TWA position. Results: Compared to MidV-CT, the MidP-CT improved by: 1) clearly reduced frequency of irregular motion artefacts, 2) increased signal to noise ratio and 3) CE increased to about 120 HU in vessels at the levels of tumour. Especially for tumours close to the thoracic wall, tumour amplitudes measured with deformable registration appeared slightly smaller than with the rigid registration method (sliding tissue effects).
T UESDAY, M AY 10, 2011
P ROFFERED PAPER
S 97
scan construction from 4D-respiration-correlated CT-scans. IJROBP 2006; 65:1560-1571. 249 oral MV IMAGE-BASED DYNAMIC MLC TRACKING OF A NITI STENT IN PIG LUNGS ON A LINEAR ACCELERATOR P. R. Poulsen1 , J. Carl2 , J. Nielsen2 , M. S. Nielsen2 , J. B. Thomsen2 , H. K. Jensen3 , B. Kjærgaard4 , P. R. Zepernick4 , E. Worm1 , W. Fledelius1 , B. Cho5 , A. Sawant6 , D. Ruan7 , P. Keall8 9 1 A ARHUS U NIVERSITY H OSPITAL, Department of Oncology, Aarhus C, Denmark 2 A ALBORG H OSPITAL , U NIVERSITY OF A ARHUS, Department of Medical Physics, Aarhus C, Denmark 3 A ALBORG H OSPITAL , U NIVERSITY OF A ARHUS, Department of Pulmonary Medicine, Aarhus C, Denmark 4 A ALBORG H OSPITAL , U NIVERSITY OF A ARHUS, Department of Thoracic Surgery, Aarhus C, Denmark 5 A SAN M EDICAL C ENTER, Department of Radiation Oncology, Seoul, Korea Republic of 6 UT S OUTHWESTERN M EDICAL C ENTER, Dallas, USA 7 U NIVERSITY OF C ALIFORNIA , L OS A NGELES, Department of Radiation Oncology, Los Angeles, USA 8 S TANFORD C ANCER C ENTER, Department of Radiation Oncology, Stanford, USA 9 T HE U NIVERSITY OF S YDNEY, Sydney Medical School, Sydney, Australia Purpose: Intrafraction tumor motion is a major challenge for precise radiotherapy of lung cancer. Dynamic MLC (DMLC) tracking is an attractive option for motion compensation and has therefore been the subject of several recent phantom studies. This work presents the first demonstration of in vivo DMLC tracking in a mammal and investigates the accuracy and potential limitations of the tracking. Materials: A thermo-expandable NiTi stent designed for kV x-ray visualization of lung lesions was inserted into the bronchia of three Gttingen minipigs. A 4DCT scan was used for planning of a 5-field conformal treatment with 6 cm circular MLC apertures. A 22.5 Gy single fraction treatment was delivered to the anaesthetized pigs. The peak-to-peak stent motion was 3-8 mm with breathing periods of 1.2-4 seconds. Prior to treatment a pair of x-ray images was used for image guided setup based on the stent. Before each field delivery about ten MV images were acquired. A small area centered at the stent was selected and used as a template for stent segmentation in continuous MV images acquired at 7.5 Hz during treatment. The segmented stent position was used for continuous adaptation of the MLC aperture (Fig. 1). A prediction algorithm accounted for the 400 ms tracking system latency. Offline, the tracking error in beam’s eye view of the MV beam was calculated for each MV image as the difference between the MLC aperture center and the segmented stent position. The standard deviations of the systematic error Σ and the random error σ were determined and compared with the would-be errors for a non-tracking treatment with pre-treatment image guided setup. The pigs were sacrificed after the treatment while still anaesthetized. The study was approved by The Animal Ethics Council under the Danish Justice Department.
Figure 1: Overview of image quality for CT-slice in mediastinum: MidV CT (top), a MidP CT (middle) and MidP BHCT (bottom) at identical level and window settings. Conclusions: Deforming all image sets derived from clinical 4D-CT-scans to a MidP-CT is a robust technique and ready to be introduced in the clinic. Image quality improves in reduced artefacts and noise. This approach also facilitates application of CE in 4D workflows. Thus, more accurate and reproducible delineation of the tumour and involved lymph nodes is expected. Wolthaus JW, Schneider C, Sonke JJ, van Herk M et al. Mid-ventilation CT
Results: Reliable stent segmentation was obtained for 11 out of 15 fields. Segmentation failures occurred when the image contrast was dominated by overlapping anatomical structures (ribs, diaphragm) rather than by the stent, which was designed for kV rather than MV x-ray visibility. For the 11 fields with reliable segmentation Σ was 0.5 mm / 0.4 mm in the two imager directions, while σ was 0.5 mm / 1.1 mm. Without tracking Σ and σ would have been 1.7 mm / 1.4 mm and 0.8 mm / 1.4 mm, respectively. For tracking, the main contributors to Σ were gantry sag and positioning reproducibility of the MV imager, which could easily be accounted for by an automatic online correction. The main contributor to σ was the latency (400 ms), which could be substantially reduced with direct image buffer access. Conclusions: For the first time, in vivo DMLC tracking has been performed on a linear accelerator demonstrating the potential for improved targeting accuracy. The study mimicked the envisioned patient workflow of future patient
P ROFFERED PAPER
T UESDAY, M AY 10, 2011
248 oral 4DCT – HIGH QUALITY IMAGING AND CONTRAST ENHANCEMENT FOR 3D RADIOTHERAPY TREATMENT PLANNING C. Schneider1 , J. Nijkamp1 , S. Rit1 , M. van Herk1 , J. J. Sonke1 1 T HE N ETHERLANDS C ANCER I NSTITUTE - A NTONI VAN L EEUWENHOEK H OSPITAL, Amsterdam, Netherlands
Conclusions: The visualization of actual dose region delivered in a treatment has been available by means of VMAT treatment. The VMAT-CT superimposed on in-VMAT kV-CBCT could be a visual verification of the actual treatment in the high-precision treatment era. 247 oral THE USE OF A GLOBAL RESPIRATORY MOTION MODEL FOR REAL TIME 4D RADIOTHERAPY APPLICATIONS H. Fayad1 , T. Pan2 , D. Visvikis1 1
INSERM U650 L ATIM, Brest, France MD A NDERSON C ANCER C ENTER, Department of physical imaging, Houston, USA 2
Purpose: Respiratory motion modeling is a key aspect for improving radiation therapy, with the objective of delivering less dose to the normal tissues and higher dose to the tumor during the breathing motion cycle. The main idea of this work is the use a generic respiratory motion model to predict in real-time the respiratory motion of the tumor and the adjacent anatomy. Materials: The generic respiratory motion model used in this work is based on principal component analysis. Only two static CT images (end-inspiration, end-expiration) in combination with respiratory synchronized patient’s body surface are needed to adapt the model on a given patient anatomy. This generic model relates the patient’s surface acquired by an external camera and the internal motion calculated using a non rigid registration algorithm. The resulting patient adapted model is subsequently used to generate in realtime the motion of the tumor and the adjacent anatomy. In this study 6 patients were used to create the model and 4 other patients were used in the validation step. The intrinsic model efficiency was tested by comparing model generated 4D-CT volumes and their corresponding acquired 4D-CT volumes. The model based motion information is subsequently used to improve planning and delivery of 4D-IMRT by accounting for 3D tumor (Suh et al 2009) as well as normal tissue motion. The whole schema of the procedure is shown in figure 1.
Results: Good correlation (r=0.93±0.4) and expert validation (1.9mm±0.6mm: based on expert anatomical landmarks identification) was found between the model generated series and the acquired 4D-CT volumes. Preliminary results indicate that including the normal tissue motion provided by our motion model during the delivery of the 4D-IMRT is better in terms of dose delivery optimization than simply accounting for tumor motion. Conclusions: The use of a global respiratory motion model, which may be adapted on a specific patient, for radiotherapy applications avoids unnecessary 4D-CT acquisitions. Accounting for normal tissue motion helps to optimize 4D dosimetry compared to considering the tumor motion only.
Purpose: Planning of radiotherapy for moving tumours in lung or abdomen can be based on respiration-correlated CT-imaging of the tumour in a time weighted averaged (TWA) position. Locating the closest bin in the respiration period corresponding to the TWA position (MidV-CT) has drawbacks : irregular breathing causes artifacts, while signal to noise ratios and contrast enhancement (CE) are poorer than in standard 3D-CT. These drawbacks can be overcome by deforming all 4D image data available to a TWA-3D scan to a Mid-Position (MidP)-CT, as proposed by Wolthaus et al [1]. The purpose of this study is to evaluate the effect of this method on image quality. Materials: For evaluation of the new technique, 10 patients were selected with a clearly visible tumour in the periphery of the lung. For all patients, a 4D-CT and an expiration breath-hold (BH) CT were acquired. For 5 of these patients, contrast was given during BH-CT. The 4D-CT-scans were reconstructed in 10 bins over the respiration cycle.The resulting 10 image sets were deformable registered, applying an iterative multi-scale phase-based optical flow estimation procedure. For each patient 2 different registrations were performed. First, the image set at maximum expiration was used as reference scan to which all other image sets were registered. Next, the average Displacement Vector Field (DVF) over all image sets was subtracted from each DVF yielding all deformations with respect to the TWA position. These DVF were applied to deform all image sets to the TWA position. Finally, the MidP-CT was calculated as median grey-value for each pixel over all 10 image sets. In the second procedure, an additional DVF from the BH-CT to the TWA position was calculated. With the latter, a MidP-BH-CT was calculated by deforming the BH-CT to the TWA position. Results: Compared to MidV-CT, the MidP-CT improved by: 1) clearly reduced frequency of irregular motion artefacts, 2) increased signal to noise ratio and 3) CE increased to about 120 HU in vessels at the levels of tumour. Especially for tumours close to the thoracic wall, tumour amplitudes measured with deformable registration appeared slightly smaller than with the rigid registration method (sliding tissue effects).
T UESDAY, M AY 10, 2011
P ROFFERED PAPER
S 97
scan construction from 4D-respiration-correlated CT-scans. IJROBP 2006; 65:1560-1571. 249 oral MV IMAGE-BASED DYNAMIC MLC TRACKING OF A NITI STENT IN PIG LUNGS ON A LINEAR ACCELERATOR P. R. Poulsen1 , J. Carl2 , J. Nielsen2 , M. S. Nielsen2 , J. B. Thomsen2 , H. K. Jensen3 , B. Kjærgaard4 , P. R. Zepernick4 , E. Worm1 , W. Fledelius1 , B. Cho5 , A. Sawant6 , D. Ruan7 , P. Keall8 9 1 A ARHUS U NIVERSITY H OSPITAL, Department of Oncology, Aarhus C, Denmark 2 A ALBORG H OSPITAL , U NIVERSITY OF A ARHUS, Department of Medical Physics, Aarhus C, Denmark 3 A ALBORG H OSPITAL , U NIVERSITY OF A ARHUS, Department of Pulmonary Medicine, Aarhus C, Denmark 4 A ALBORG H OSPITAL , U NIVERSITY OF A ARHUS, Department of Thoracic Surgery, Aarhus C, Denmark 5 A SAN M EDICAL C ENTER, Department of Radiation Oncology, Seoul, Korea Republic of 6 UT S OUTHWESTERN M EDICAL C ENTER, Dallas, USA 7 U NIVERSITY OF C ALIFORNIA , L OS A NGELES, Department of Radiation Oncology, Los Angeles, USA 8 S TANFORD C ANCER C ENTER, Department of Radiation Oncology, Stanford, USA 9 T HE U NIVERSITY OF S YDNEY, Sydney Medical School, Sydney, Australia Purpose: Intrafraction tumor motion is a major challenge for precise radiotherapy of lung cancer. Dynamic MLC (DMLC) tracking is an attractive option for motion compensation and has therefore been the subject of several recent phantom studies. This work presents the first demonstration of in vivo DMLC tracking in a mammal and investigates the accuracy and potential limitations of the tracking. Materials: A thermo-expandable NiTi stent designed for kV x-ray visualization of lung lesions was inserted into the bronchia of three Gttingen minipigs. A 4DCT scan was used for planning of a 5-field conformal treatment with 6 cm circular MLC apertures. A 22.5 Gy single fraction treatment was delivered to the anaesthetized pigs. The peak-to-peak stent motion was 3-8 mm with breathing periods of 1.2-4 seconds. Prior to treatment a pair of x-ray images was used for image guided setup based on the stent. Before each field delivery about ten MV images were acquired. A small area centered at the stent was selected and used as a template for stent segmentation in continuous MV images acquired at 7.5 Hz during treatment. The segmented stent position was used for continuous adaptation of the MLC aperture (Fig. 1). A prediction algorithm accounted for the 400 ms tracking system latency. Offline, the tracking error in beam’s eye view of the MV beam was calculated for each MV image as the difference between the MLC aperture center and the segmented stent position. The standard deviations of the systematic error Σ and the random error σ were determined and compared with the would-be errors for a non-tracking treatment with pre-treatment image guided setup. The pigs were sacrificed after the treatment while still anaesthetized. The study was approved by The Animal Ethics Council under the Danish Justice Department.
Figure 1: Overview of image quality for CT-slice in mediastinum: MidV CT (top), a MidP CT (middle) and MidP BHCT (bottom) at identical level and window settings. Conclusions: Deforming all image sets derived from clinical 4D-CT-scans to a MidP-CT is a robust technique and ready to be introduced in the clinic. Image quality improves in reduced artefacts and noise. This approach also facilitates application of CE in 4D workflows. Thus, more accurate and reproducible delineation of the tumour and involved lymph nodes is expected. Wolthaus JW, Schneider C, Sonke JJ, van Herk M et al. Mid-ventilation CT
Results: Reliable stent segmentation was obtained for 11 out of 15 fields. Segmentation failures occurred when the image contrast was dominated by overlapping anatomical structures (ribs, diaphragm) rather than by the stent, which was designed for kV rather than MV x-ray visibility. For the 11 fields with reliable segmentation Σ was 0.5 mm / 0.4 mm in the two imager directions, while σ was 0.5 mm / 1.1 mm. Without tracking Σ and σ would have been 1.7 mm / 1.4 mm and 0.8 mm / 1.4 mm, respectively. For tracking, the main contributors to Σ were gantry sag and positioning reproducibility of the MV imager, which could easily be accounted for by an automatic online correction. The main contributor to σ was the latency (400 ms), which could be substantially reduced with direct image buffer access. Conclusions: For the first time, in vivo DMLC tracking has been performed on a linear accelerator demonstrating the potential for improved targeting accuracy. The study mimicked the envisioned patient workflow of future patient