Tumor motion during liver SBRT - projects at Aarhus University Hospital - Per Poulsen, Esben Worm, Walther Fledelius, Morten Høyer Aarhus University Hospital, Denmark
SBRT: Stereotactic Body Radiation Therapy Few fractions (1-5) with high daily doses (12-25 Gy) Steep dose gradients High precision needed Traditionally an external stereotactic coordinate system is used for precise targeting but this tends to be replaced by IGRT
IGRT for liver SBRT Liver tumors are very difficult to image in-room [contrast required] Tumor position surrogates: Bony anatomy, diaphragm, implanted markers Implanted markers are more accurate than anatomical land marks [if marker is less than ~8 cm from tumor ( * ) ] ( * ) Seppenwoolde PMB 2011
Outline Liver SBRT planning and setup Monitoring intrafraction tumor motion Tumor motion characteristics Trajectory based patient setup Dose reconstruction Simulated Dynamic MLC tracking Real-time marker segmentation
Liver SBRT planning and setup Marker implantation Fixation CT scanning Target definition Treatment planning Setup 2-3 gold markers. 1 mm x 3 mm Implanted percutaneously guided by ultrasound Ideally close to and surrounding the tumor Often limited implantation directions
Liver SBRT planning and setup Marker implantation Fixation CT scanning Target definition Treatment planning Setup Stereotactic Body Frame (Elekta) If possible: Photo from Berbeco et al. IJORBP 2007 Abdominal compression to reduce breathing motion
Liver SBRT planning and setup Marker implantation Fixation CT scanning Target definition Treatment planning Setup CT1: Expiration breath-hold with contrast [optimal contrast timing window ~15 seconds] Good contrast timing - and a bit later
Liver SBRT planning and setup Marker implantation Fixation CT scanning Target definition Treatment planning Setup CT2: 4DCT c
Liver SBRT planning and setup Marker implantation Fixation CT scanning Target definition Treatment planning Setup 1. Match breath-hold CT and 4DCT mid-ventilation based on markers 2. Delineate GTV in 4DCT mid-vent assisted by contrast enhanced breath-hold CT 3. CTV = GTV 4. PTV = CTV + 5mm 5mm 10mm
Liver SBRT planning and setup Marker implantation Fixation CT scanning Target definition Treatment planning Setup Risk organs: Healthy liver (Liver CTV) Spinal cord (with 10 mm PRV margin) Esophagus (with 10 mm PRV margin) Stomach, duodenum, bowel, kidneys, heart
Liver SBRT planning and setup Marker implantation Fixation CT scanning Target definition Treatment planning Setup Most often conformal plan with IM segments IMRT or VMAT may be used for complicated cases CTV covered with 95% dose, PTV with 67% dose
Liver SBRT planning and setup Marker implantation Fixation CT scanning Target definition Treatment planning Setup Risk adapted dose prescription for metastasis: Use highest dose among 3 12.5Gy, 3 13.75Gy, 3 15Gy and 3 16.75Gy that fulfills constraints:
Liver SBRT planning and setup Marker implantation Fixation CT scanning Target definition Treatment planning Setup 1. Laser guided positioning of patient in Stereotactic Body Frame 2. Room laser guided positioning of Stereotactic Body Frame relative to linac isocenter 3. Cone-beam CT with match to planning CT (=midventilation 4DCT phase): a. Automatic match on bony anatomy b. Manual match on markers (max 10 mm shift)
Outline Liver SBRT planning and setup Monitoring intrafraction tumor motion Tumor motion characteristics Trajectory based patient setup Dose reconstruction Simulated Dynamic MLC tracking Real-time marker segmentation
Monitoring intrafraction liver tumor motion PhD project, Esben Worm, started June 2010 Aim: Detailed characterization of tumor motion in liver SBRT Method: Monitoring liver marker motion during the entire treatment delivery NB: Esben to visit U Sydney in April-May 2012
Monitoring intrafraction liver tumor motion Planning: 3 x treatment: 4DCT time Record external motion (RPM) + kv/mv images during the entire treatment RPM: CBCT1 (setup) CBCT2 (control) Field 1 MV+kV Field 2 MV+kV Field n MV+kV CBCT3 time
Intra-treatment imaging kv, 5 Hz RPM, 25 Hz MV, 7.5-7.8 Hz
Example 1: Intra-treatment motion MV kv
Example 2: Motion in MV images DRR Mid-ventilation 7.5 Hz portal imaging Cranio-caudal geometrical error
Example 2: Motion in MV images Fraction 1 Fraction 2 Fraction 3 Note the large day-to-day variation in marker positions
Pre- and post-treatment imaging RPM CBCT: 1. Acquire ~650 projection images in ~1 minute during a full gantry rotation 2. Reconstruct 3D image from projection images
Examples: Liver CBCT scans: Motion artifacts Example 1: Diaphragm 15 mm CBCT CT Example 2: Gold marker (3 mm) 12 mm
Examples: Liver CBCT projections 1 CBCT projection (out of 629) Marker motion Marker motion
2D 3D?
f 3 f 2 Probability-Based 3D Trajectory Estimation Image 1 f 1 Image 3 Target is known to be located on a sequence of ray lines
Assume 3D Gaussian Probability Density f 3 f 2 Image 1 Gaussian PDF f 1 Image 3 Fit 3D Gaussian PDF to observed projections
Estimate 3D target position from PDF f 3 f 2 Image 1 Gaussian PDF f 1 Image 3 Select most likely position along each ray line
Result: 3D trajectory 1 Image 1 3 2 Image 3
Single-imager trajectory estimation Accurate: 3D RMS error < 1mm in 99.1% of cases for lung/abdomen and in 99.7% of cases for prostate Can be used in real-time with modest loss in accuracy as compared to retrospective ( Tracking) Poulsen et al: IJROBP 2008, PMB 2009
Example: Motion during CBCT acquisition External RPM signal External Internal Internal tumor motion (from CBCT projections)
Amending missing intra-treatment motion data Amend with single imager method when: kv images are missing [not recorded, marker not segmentable due to MV scatter] MV images are missing [not recorded, marker outside field, IMRT, VMAT] Amend with RPM signal when: Both MV and kv are momentarily missing [less than one breathing cycle at a time for MV] [only a few breathing cycles for kv]
Monitoring intrafraction liver tumor motion Planning: 3 x treatment: 4DCT time Synchronized 1D external abdominal motion and 3D internal tumor surrogate motion during the entire treatment RPM: CBCT1 (setup) CBCT2 (control) Field 1 MV+kV Field 2 MV+kV Field n MV+kV CBCT3 time
Fraction 1 Fraction 2 Fraction 3 Example: 3D trajectories for one patient
Outline Liver SBRT planning and setup Monitoring intrafraction tumor motion Tumor motion characteristics Trajectory based patient setup Dose reconstruction Simulated Dynamic MLC tracking Real-time marker segmentation
Status and perspectives of liver motion study 15 patients recruited Allows addressing a wide range of questions about: Tumor motion characteristics (magnitude, directionality, stability, predictability, correlation with 4DCT, correlation with RPM, etc, on all relevant time scales) Use of adaptive margins, gating & tracking based on limited information (e.g kv alone, RPM+kV...) Delivered tumor dose distribution (in actual treatment, in simulated gating or tracking treatment, etc.)
Motion (mm) CBCT AP (mm) M CB -26-28 -28 Example: Tumor trajectory during CBCT scan -30-10 0 10 20 30 40 50 60 70 Time (s) -26-30 76 78 80 82 84 86 88 RPM AP (mm) 46 44 42 CBCT AP RPM 46 44 42 40 38 36 34 40 38 36 34 RPM vs CBCT AP Pearsons R = 0.9649 32-10 0 10 20 30 40 50 60 70 Time (s) 32 76 78 80 82 84 86 88 RPM AP (mm) Example: Correlation between internal motion (black) and external motion (blue). Could be used for proper gating window setting based on CBCT.
LR (mm) Example: Variation in breathing motion -1-2 -3 Max respiration Min respiration 4 2 0-2 CC (mm) -4-6 -8 Smallest and largest respiratory cycle for a patient during beam-on -10-12 -14 0 2 4 AP (mm) 6
field5 field5 field5 Intra-treatment time field4 field4 field4 field3 field3 field3 field2 field2 field2 field1 field1 field1 CBCT CBCT CBCT Example: 3D PDF evolvement for single fraction 1 0.5 1 0.5 1.7 mm 1 0.5 0 1 0.5 0 1 0.5 0 1 0.5 0 1 0.5 0 1 0.5 0 1 0.5 0 1 0.5 0 1 0.5 0 1 0.5 0 1 0.5 0 1 0.5 0 1 0.5 0 1 0.5 0-2 0 2 4 AP Position (mm) 0 1 0.5 0-5 0 5 CC Position (mm) 0 1 0.5 0-4 -3-2 -1 LR Position (mm)
Example: Intrafraction mean position drift (9 pat) CBCT First field First Field Last field CBCT Last field
Example: Motion directionality, Pat1 Sagittal view
Example: Motion directionality, 9 patients Spherical view
Outline Liver SBRT planning and setup Monitoring intrafraction tumor motion Tumor motion characteristics Trajectory based patient setup Dose reconstruction Simulated Dynamic MLC tracking Real-time marker segmentation
CBCT based setup in liver SBRT Respiratory motion during CBCT scannning Motion blurring of markers Difficult to find the mean marker position for CBCT planct match Trajectory based setup would be better
Trajectory based setup in liver SBRT CC (mm) LR (mm) AP (mm) Segment 2D trajectory 2 0 3D trajectory -2 0 10 20 30 40 50 60 0-5 Mean marker position can be found unambiguously from the trajectory -10-15 0 10 20 30 40 50 60 10 5 0 Estimated motion Estimated mean CBCT marker match CBCT bony match 0 10 20 30 40 50 60 Time (seconds) Esben Worm IJROPB 2012
Online trajectory based setup Custom made semiautomatic segmentation program Esben Worm IJROPB 2012
Online trajectory based setup Custom made semiautomatic segmentation program Esben Worm IJROPB 2012
Trajectory based setup in liver SBRT Status: Used on-line for 2 patients Tested off-line (post-treatment) for 5 other patients Next: Go from hand-held semi-automatic method to fully automatic method integrated into the clinical workflow
Outline Liver SBRT planning and setup Monitoring intrafraction tumor motion Tumor motion characteristics Trajectory based patient setup Dose reconstruction Simulated Dynamic MLC tracking Real-time marker segmentation
Dose reconstruction in DMLC tracking CIRS lung phantom with 1D moving rod (CC direction) Solid water tumor insert (24mm high, 25mm Ø) Sandwiched vertical Gafchromic EBT film Left lateral conformal treatment field Static or sinusoidal motion (20mm peak-to-peak, T = 4 sec) MV image-based tracking using an embedded gold marker Mai Lykkegaard Schmidt, Lone Hoffmann
Static Measured dose distributions Motion Tracking Static Motion Tracking
Dose reconstruction I: Conformal, no tracking TPS Experiment Dicom plan Target motion Motion mimicking Dicom plan Dose reconstruction in TPS Motion mimicking treatment plan Divide target motion into 1 mm position bins Create a sub-beam for each position bin Shift the sub-beam isocenter to mimick target motion Method equivalent to Berbeco, Med Phys 2008
Dose reconstruction II: Tracking TPS Experiment Experiment Dicom plan Target motion MLC positions Motion mimicking Dicom plan Dose reconstruction in TPS The sub-beam for each target position bin is now a step-and-shoot IMRT beam
Dose reconstruction characteristics Includes effects of: MLC shape changes 3D target translations Interplay effects Physical path length changes Does not include: Radiological path length changes Target deformations Target motion relative to the marker
Reconstructed and measured dose distributions Reconstructed Reconstructed Measured Measured
Reconstructed and measured dose distributions Comparison of measured and reconstructed doses: 2%/2mm gamma pass rates: 99.5-99.9% Mai Lykkegaard Schmidt Lone Hoffmann
Reconstructed and measured dose distributions Comparison of measured and reconstructed doses: 2%/2mm gamma pass rates: 99.5-99.9% Motion parallel to MLC GTV DVHs Motion perpendicular to MLC GTV DVHs
Outline Liver SBRT planning and setup Monitoring intrafraction tumor motion Tumor motion characteristics Trajectory based patient setup Dose reconstruction Simulated Dynamic MLC tracking Real-time marker segmentation
Simulated MV based DMLC tracking for liver SBRT Background Detailed knowledge of prototype DMLC tracking system (latencies, prediction errors, refitting of MLC) Detailed measurements of MLC response to leaf motion requests Examples of leaf adaptation dynamics measured on millisecond time scale
Simulated MV based DMLC tracking for liver SBRT Simulate MLC motion by modeling the effects of MV tracking latency (290 ms) Prediction for latency compensation Refitting of leaves to the estimated target position Re-adjustment of leaves to the refitted positions
Simulated MV based DMLC tracking for liver SBRT Assume correct real-time marker segmentation in all MV images 2D DMLC tracking Prediction starts after 8 seconds treatment Assume marker visibility in all MV images of main fields No tracking for IM segments Similar to pig tracking experiment, Poulsen IJROBP 2011
Results example: Patient #2/5: MV images MV. Pat02. Fx1. Field1 Beam s eye view error Green: Gold Marker Red: CTV
Results example: Patient #2/5: MV images MV BEV error for all fields at fraction 1 Field 1 Field 2 Field 3 Field 4 Field 5 Mean CC error: -4.1 mm
Results example: Patient #2/5: Tracking Simulation MV. Pat02. Fx1. Field1 MLC motion in simulated DMLC tracking treatment guided by these MV images The residual errors with tracking tend to be random Green: Gold Marker Red: CTV
Dose reconstruction: As Treated sagittal view BEV 2/299 MV images 2.572 MU 16/299 MV images 20.576 MU Original iso-c
Dose reconstruction: Simulated tracking sagittal view BEV 2/299 MV images 2.572 MU Step&shoot IMRT With 2 segments 16/299 MV images 20.576 MU Original iso-c Step&shoot IMRT With 16 segments
Motion including treatment plan: All subfields As treated Simulated tracking Motion speed doubled
Reconstructed dose distributions Dosewash: >95% dose Red = CTV Patient 2 Fraction 1
Reconstructed dose distribution DVH for CTV Fraction 1 Delivered CTV dose at first fraction: D 95 = 7%
Reconstructed dose distribution DVH for CTV Fractions 1-3 Delivered CTV dose at all three fractions: D 95 = 7%, D 95 = 5.6%, D 95 = 2.5% Tracking would have restored the planned CTV dose
Results: Minimum dose to 95% of the CTV Reduction in D 95 (%-point) As treated Simulated tracking Mean Maximum Mean Maximum 2.0 7.0 0.3 0.5
Results: Minimum dose to 95% of the CTV? Reduction in D 95 (%-point) As treated Simulated tracking Mean Maximum Mean Maximum 2.0 7.0 0.3 0.5
Increased D 95 with motion?? Planned (static) Treated (w/motion) Color dose wash: 97.3% ( D 95 ) Motion blurring increases V 97.3%
D 95 correlates with mean geometrical error p < 0.001.
D 95 correlates with mean geometrical error p < 0.001.
Conclusion for DMLC tracking simulations Tumor motion can deteriorate the target dose in liver SBRT even with daily image-guided localization DMLC tracking can restore the planned dose DMLC tracking made no difference for 3 out of 5 investigated treatment courses The tracking would be a safety net for the outliers and would not harm the inliers (no imaging dose).
DMLC tracking simulations: Possible usage Dosimetry evaluation Dosimetric gain of tracking Tracking QA: Test robustness of a given DMLC plan against a wide range of tumor motions Development of new tracking strategies MRI, multiple targets, deformations,.
Outline Liver SBRT planning and setup Monitoring intrafraction tumor motion Tumor motion characteristics Trajectory based patient setup Dose reconstruction Simulated Dynamic MLC tracking Real-time marker segmentation
Multiple marker segmentation Challenges for segmentation: Changing shape with projection angle Marker overlap, trace entanglement Poor contrast in lateral pelvic images Motion of markers relative to each other
Multiple marker segmentation Aim: Fully automatic No prior knowledge For both abdomen/thorax and pelvis Fledelius MedPhys 2011
Step 1: Identify marker candidates in each frame Single frame of pelvic CBCT scan. Four markers present
Step 1: Identify marker candidates in each frame Marker candidates identified in the frame (blobs)
Step 2: Identify trace candidates - by following marker candidates in subsequent frames
Step 3: Reject trace candidates - that are inconsistent with a 3D marker position
Step 4: Construct 3D marker constellation model - by using the most reliable remaining traces
Step 5: Auto-segment i all images - using projections of the 3D model as templates 3D model Anterior view Anterior image Project
Step 5: Auto-segment i all images - using projections of the 3D model as templates 3D model Lateral view Lateral image Project
Step 5: Auto-segment i all images - using projections of the 3D model as templates Segmentations failures: In 0.2 % of 42000 prostate cases In 0.1 % of 40000 liver cases Fledelius MedPhys 2011
Real-time auto-segmentation: Liver SBRT Pre-treatment CBCT scan: 3D marker constellation model Clouds of visited 3D positions Fledelius ESTRO 2012
Real-time auto-segmentation: Liver SBRT Projection image Templates Search areas Fledelius ESTRO 2012
Real-time auto-segmentation: Liver SBRT In CBCT projections In intra-treatment kv Fledelius ESTRO 2012
Summary 1. Liver SBRT planning: Risk adapted dose prescription 2. Intrafraction motion monitoring: OBI/kV/MV/RPM 3. Tumor motion characteristics: magnitude, direction, stability, correlation. 4. Trajectory based patient setup: Improves accuracy 5. Dose reconstruction: Generalized to any treatment type. 6. Simulated Dynamic MLC tracking: Dosimetric gain with tracking? 7. Multiple marker segmentation: Going real-time.
Thank you Esben Worm Walther Fledelius Thomas Ravkilde Morten Høyer Mai Lykkegaard Schmidt Lone Hoffmann Paul Keall & research group, Sydney Varian, Varian ilab