Multimodality Imaging in Radiation Oncology: Imaging vs. Imagining Multimodality & 4D Image Registration: Mehau Kulyk /spL Methods & By 9:25 AM, you will …. Understand the basic mechanics of multimodality and 4D image registration techniques Clinical Use Understand the different techniques used to combine, display and interact with multimodality and 4D image and dose data Marc L Kessler, PhD The University of Michigan Understand the clinical use and limitations of these techniques for Tx planning, Tx delivery and plan adaptation Look here! By 9:25 AM, you will …. ing n pe d” … p ha hoo used Understand the s different techniques i - with to combine,tdisplay and interact e a h multimodality -4Dtimage and dose data wh and r … Understand the use and limitations declinical n of these techniques for Tx planning, Tx u “ delivery and plan adaptation Understand the basic mechanics of multimodality and 4D image registration techniques Acknowledgements! Many wonderful people have contributed material for this presentation! Marc L Kessler, PhD 1 Multimodality Imaging in Radiation Oncology: Imaging vs. Imagining Outline Motivation Precision radiation therapy requires accurate delineation of the tumor and normal tissues in the planning phase and accurate localization of these structures during the delivery phase Motivation Mechanics ! Clinical Use …with the aid of imaging Gregoire Gregoire // St-Luc St-Luc Motivation Multimodality Targeting entire Optimization of the radiotherapy process requires that we anticipate, measure & adapt to changes in the patient Imaging Planning ? Delivery on-line off-line Imaging Physics X-ray CT Anatomy MRI Physiology Nuc Med We now have many cameras available … which provide complementary data! Marc L Kessler, PhD 2 Multimodality Imaging in Radiation Oncology: Imaging vs. Imagining Repeat Imaging 4-D Imaging Balter Balter // UM UM ? ? … assess motion Normal Tissues Target Volumes Dawson Dawson // PMH PMH Image Guided Treatment Varian OBI™ Elekta Synergy™ Siemens PRIMATOM™ TomoTherapy Hi-Art™ ViewRay Renaissance™ Resonant Restitu™ The Big Picture CT Tx Plan 3D Dose MR NM “Adapting” Patient Model US 3D Dose Day n Marc L Kessler, PhD Portal Images CBCT 1…n US 4D CBCT 3 Multimodality Imaging in Radiation Oncology: Imaging vs. Imagining The Goal 7/26/2007 Ideally, we would like to have a time dependent vector of information for every “point” in an anatomic object image information (MR, CT, NM, … ) physiologic information (τ ) anatomic label information dose information … with time stamp ! The Goal Past / Present PET MR Registration CT Segmentation CT+ MR + NM + Dose (τττ) Scalar Data 4-D Vectors Marc L Kessler, PhD 4 Multimodality Imaging in Radiation Oncology: Imaging vs. Imagining Present / Future Outline Motivation Registration Mechanics ! Segmentation Clinical Examples Mechanics Transformation … determine the geometric transformation that maps corresponding points from one image series to another … determine the geometric transformation that maps corresponding points from one image series to another Form of the transformation T … from rigid to fully freeform Number of degrees of freedoms β … from 3 to 3 x N * **N XB = T ( XA , { ß }) (x,y,z) coordinates of a point in Series B (x,y,z) coordinates of a point in Series A = number of voxels Marc L Kessler, PhD 5 Multimodality Imaging in Radiation Oncology: Imaging vs. Imagining What is T ? Degrees of Freedom PET/CT MR - CT 4D CT Rigid / Affine Global, regional, or piecewise Full 3D / 4D Deformation Parametric models Free-form models None ? Few Many www.gnome.org www.gnome.org What is T ? Affine Transformations Rigid / Affine Global, regional, or piecewise xB = A xA + b y = m x+ b Study A A A Square Square Study B (up to 12 DOF) … in 3D Translation Translation Rotation Rotation Scaling Scaling Shearing Shearing 3 3 3 3 Parallel lines stay parallel ! Marc L Kessler, PhD 6 Multimodality Imaging in Radiation Oncology: Imaging vs. Imagining www.gnome.org www.gnome.org Affine Transformations Study A 6 DOF Translation Translation Study A Affine Transformations Study A A A Square Square Rotation Rotation A A Square Square Study B Scaling Scaling Shearing Shearing Parallel lines stay parallel ! non- Affine www.gnome.org www.gnome.org Transformations Study B Parallel lines don’t stay parallel! Study B Translation Translation Rotation Rotation Scaling Scaling Shearing Shearing 3 or 4 DOF Parallel lines stay parallel ! non- Affine Study A Transformations Study B XB = T ( XA , { ß }) Marc L Kessler, PhD 7 Multimodality Imaging in Radiation Oncology: Imaging vs. Imagining non- Affine Study A Transformations non- Affine Study B Transformations Study A Study B Transformation parameters to apply to a particular point depends on the location of the point ! XB = T ( XA , { ß(XA) }) non- Affine Transformations Balter Balter // UM UM XB = T ( XA , { ß(XA) }) Full 3D / 4D Deformation … up to 3 x N (DICOM!) phase dependent ? XB = T ( XA , { ß(XA, φ )}) Parametric Marc L Kessler, PhD Freeform 8 Multimodality Imaging in Radiation Oncology: Imaging vs. Imagining Full 3D / 4D Deformation Parametric Full 3D / 4D Deformation Each have some distinct properties … local Various splines ( TPS , B-splines ) B-Splines Other basis functions Thin-Plate splines … global Freeform Finite element models Flow models ( optical, viscous ) Full 3D / 4D Deformation Warp Space / … Drag Objects Warp Objects / … Drag Space Brock / PMH Parametric Finite element … bio-mechanical Intensity flow … image forces ( mono-modality ) How Do We Compute { β } ? 1 Construct a metric that measures the mismatch (or similarity) between a pair of datasets 2 Apply an optimization algorithm to determine the parameters (DOF) that minimize (maximize) this metric Freeform Marc L Kessler, PhD 9 Multimodality Imaging in Radiation Oncology: Imaging vs. Imagining How Do We Compute { β } ? {β} Registration Metrics ? Geometry-Based Metrics Point Matching ? Least Squares Surface Matching Chamfer Matching Geometry-based Intensity-based Marc L Kessler, PhD Σ( X B B - XAA ) 2 Σ min distance 2 … depends on image segmentation! 10 Multimodality Imaging in Radiation Oncology: Imaging vs. Imagining Intensity-Based Metrics Mono-modality Sum Squared Difference Σ(I B B How About An Example? Transformation - IAA ) 2 Rotate - Translate PET Registration Metric Multimodality Data Mutual Information Σ p(IAA, IBB) p(IAA, IBB) log p(IAA) p(IBB) CT Mutual Information Optimizer Simplex Algorithm … depends on the image characteristics! How About An Example? How About An Example? PET CT PET CT Marc L Kessler, PhD 11 Multimodality Imaging in Radiation Oncology: Imaging vs. Imagining How About An Example? Balter Balter // UM UM How About Deformations ? PET CT Multiphasic CT Data How About Deformations ? Transformation B-Splines ( multi-resolution ) Multiresolution Deformations Successively increase the resolution of the knot spacing Registration Metric Sum Squared Difference Optimizer Exhale State decent Gradient Inhale State Only small additional computation cost when increasing the number of knots. Marc L Kessler, PhD 12 Multimodality Imaging in Radiation Oncology: Imaging vs. Imagining Multiresolution Deformations Multiresolution Deformations Successively increase the resolution of the image data Successively increase the resolution of the image data ¼ Resolution ¼ Resolution Full Resolution Full Resolution 60 60 xx 48 48 mm mm 60 xx 60 4 4 xx 3 3 mm mm 4 xx 4 Coarse Fine Coarse Fine Multiresolution Deformations Multiresolution B-Splines Registration Metric vs. Iteration Multiphasic CT Data Registration Metric 2.5 2.5 Change in knot spacing 2.4 2.4 Low Res 2.3 2.3 2.2 2.2 2.1 2.1 High Res 2.0 2.0 1.9 1.9 Exhale State 1.8 1.8 0 0 20 20 40 40 60 60 80 80 100 100 120 120 140 140 160 160 180 180 Inhale State deformed Iteration Number Marc L Kessler, PhD 13 Multimodality Imaging in Radiation Oncology: Imaging vs. Imagining Ruan Ruan // UM UM Multiresolution B-Splines We Are Not Really Splines ! Multiphasic CT Data No “stiffness” information Exhale State Inhale State deformed Add Some Physics? Extracted Ribcage Exhale Deform Inhale Spatially Variant Stiffness Etotal = Esimilarity + α Estiffness intensity similarity measure tissue-dependent regularization Evol = wc(x) |det JT(x) – 1|2 dx Marc L Kessler, PhD 14 Multimodality Imaging in Radiation Oncology: Imaging vs. Imagining Ruan Ruan // UM UM “Stiffness” Weighting Using “Prior” Information wc(x) using “stiffness” information Extracted Ribcage Tissue Sliding Balter Balter // UM UM Deal with different organs individually? Exhale Deform Inhale Tissue Sliding Balter Balter // UM UM Deal with different organs individually? Marc L Kessler, PhD 15 Multimodality Imaging in Radiation Oncology: Imaging vs. Imagining Segmentation + Registration No masking Masking Brock Brock // UM UM Finite Element Modeling Exhale Exhale Inhale Inhale Ribs driven by large lung deformations Ribs not affected by lung registration Brock Brock // PMH PMH Finite Element Modeling Take into account physical tissue properties (directly) The Future ? Family of Generalized, Customizable, Patient Models … thorough segmentation is necessary Marc L Kessler, PhD 16 Multimodality Imaging in Radiation Oncology: Imaging vs. Imagining Is The Future Here Already? Is It Here Already? II have have no no commercial commercial interest interest in in this this company company www.mimvista.com www.mimvista.com …from Atlas to Individual II have have no no commercial commercial interest interest in in this this company company Thompson Thompson // UCLA UCLA …from Individuals to Atlas Brain Brain Mapping: Mapping: The The Disorders Disorders,, Academic Academic Press, Press, 1999 1999 Marc L Kessler, PhD 17 Multimodality Imaging in Radiation Oncology: Imaging vs. Imagining Meyer/UM Meyer/UM without Segment /register / average In The Meantime … with with Segment using atlas Anatomy Mapping Image Anatomy Dose Anatomy Mapping Dong Dong // MDACC MDACC Drawn Contours Simple Overlay (no transform) Planning CT “Delivery” CT … map to CT Boolean OR Use superior MR contrast for targeting Marc L Kessler, PhD 18 Multimodality Imaging in Radiation Oncology: Imaging vs. Imagining Anatomy Mapping Drawn Contours Dong Dong // MDACC MDACC Transformed and resampled Segmentation of a registration! Planning done CT w/ the aid“Delivery” CT Pouliot Pouliot // UCSF UCSF Dose Mapping / Analysis CBCT 1 Dose Mapping Dong Dong // MDACC MDACC CBCT 2 Transforming All Voxels! More Than Deformations ∆ Dose deformation weight loss resection Change in shape Dose Difference (%) >5% Increased cord dose … not just deformation! shrinkage ∆ vascular >10% Marc L Kessler, PhD 19 Multimodality Imaging in Radiation Oncology: Imaging vs. Imagining Dose Mapping Validation Dealing with volume elements that may: change shape / appear / disappear … need proper spatial re-sampling How do we know how well these registration methods perform? build phantoms and test them don’t necessarily add in a linear fashion we can know the truth! … need some sort of radiobiology provide tools to examine results exist in homogenous intensity regions we don’t know the truth! … hard to evaluate registration Validation Phantoms 1986 1986 Validation Phantoms Kashani Kashani // UM UM CT MR Marc L Kessler, PhD 20 Multimodality Imaging in Radiation Oncology: Imaging vs. Imagining Validation Tools Qualitative Tools Validation Tools Quantitative Tools Study A Color gel or wash overlay Split /dual screen displays Point 1 2 3 4 5 6 Description 2nd branch of bronchial tree 3rd branch of bronchial tree 4th branch of bronchial tree Vessel bifurcation 1 Vessel bifurcation 2 Vessel bifurcation 3 X -5.37 -5.73 -6.50 -8.12 -8.06 -10.69 Exhale Y 0.98 2.12 2.77 3.37 -1.95 2.47 Z -3.42 -5.42 -8.42 A A -9.92 -4.42 0.58 Z -2.92 -5.92 -9.42 -11.42 -3.92 1.08 Exhale' - Inhale ∆Y -0.25 -0.16 -0.11 -0.49 0.37 -0.29 ∆Z -0.44 0.09 -0.09 -0.18 0.29 0.16 0.29 0.26 X -4.62 -5.40 -6.24 A A A -8.12 A -7.67 -10.78 (x , y , z ) Study B all values in cm. Anatomic boundary overlay! Exhale' ( w/ TPS alignment ) -4.71 -0.47 -3.36 -5.35 0.58 -5.83 -6.27 0.69 -9.51 B B -8.19 0.91 -11.60 -7.27 -2.83 -3.63 -10.85 0.87 1.24 * ∆X -0.09 0.05 -0.03B B B B -0.07 0.40 -0.07 (x , y , z ) σ AAPM Task Group 132 Inhale Y -0.22 0.74 0.80 1.40 -3.20 1.16 * 0.19 Opportunities & Challenges T22 Use of Image Registration and Data Fusion Algorithms and Techniques in Radiotherapy Methods to assess the accuracy of i ng image registration andmfusion o C o ‘08 in T11 Flair Gd Diff Issues related to acceptance testing and quality assurance for image registration and fusion Marc L Kessler, PhD 21 Multimodality Imaging in Radiation Oncology: Imaging vs. Imagining More than just mechanics! Summary What Now ? Taxonomy of Registration Process Geometry Intensity Interactive Automated Affine non-Affine MR volumes mapped to CT study Summary Product Comparison Tools are now available to register and integrate image, anatomy & dose for both Tx planning and Tx delivery www.ITNonline.net These tools can be used to help build better models of the patient and to help customize and adapt therapy Work towards more standard and robust tools and validations methods (for non-rigid) situations continues Marc L Kessler, PhD 22 Multimodality Imaging in Radiation Oncology: Imaging vs. Imagining Don’t try this at home! Thank you for your time ! There’s more than one way to scan a cat www.itk.org Winter Institute of Medical Physics February 9-13, 2008 30th Year Anniversary www.utmem.edu/WIMP/ Summit County, Colorado ~ Breckenridge, Keystone, Vail, A-Basin, Copper ~ Marc L Kessler, PhD 23 Multimodality Imaging in Radiation Oncology: Imaging vs. Imagining Practical Aspects Degrees of Freedom … do whatever you can to reduce the number of degrees of freedom of the image registration problem ! PET/CT MR - CT 4D CT Using too many degrees of freedom will … increase computation time increase the likelihood of local minima likely decrease the overall accuracy None ? Degrees of Freedom PET Few Many Not Always None! CT PET / CT Hybrid GE Discovery LS T = Identity ? XPET = XCT ? PET Marc L Kessler, PhD XPET = XCT CT 24 Multimodality Imaging in Radiation Oncology: Imaging vs. Imagining JNM JNM 46:1488-96 46:1488-96 2005 2005 Not Always None! Not Always None! User Beware! CT artifact from respiration “burned in” to attenuation corrected PET Not Always None! Dawson Dawson // PMH PMH What About Using Just a Few? MR - CT phantom in head frame Most of the motion of the liver seems to be rigid or affine! … some deformation does occur though. Mechanically attach a coordinate system You still need to be careful ! Marc L Kessler, PhD 25 Multimodality Imaging in Radiation Oncology: Imaging vs. Imagining What About Using Just a Few? What About Using Just a Few? maybe ignore over a limited field-of-view maybe ignore over a limited field-of-view ? particularly poor MR CT ( Diagnostic ) ( Therapy ) anatomic-based data cropping CT registered to MR split-screen display What About Using Just a Few? What About Using Just a Few? maybe ignore over a limited field-of-view maybe ignore over a limited field-of-view anatomic-based data cropping anatomic-based data cropping CT registered to MR image-switch display Marc L Kessler, PhD CT registered to MR image-switch display 26 Multimodality Imaging in Radiation Oncology: Imaging vs. Imagining Limited Field-of-View Limited Field-of-View Rigid assumption used for regional registration Oops! Pick Your Battles Wisely! Successively increase the resolution of the image data and spline knot density 60 60 xx 60 60 xx 48 48 mm mm 4 4 xx 4 4 xx 3 3 mm mm Coarse Fine Knot Spacing Multiresolution Deformations Registration Metric vs. Iteration 2.5 2.5 Registration Metric Start Off Slow, … Speed Up! Increase in knot density 2.4 2.4 2.3 2.3 2.2 2.2 2.1 2.1 2.0 2.0 1.9 1.9 1.8 1.8 0 0 20 20 40 40 60 60 80 80 100 100 120 120 140 140 160 160 180 180 Iteration Number Marc L Kessler, PhD 27 Multimodality Imaging in Radiation Oncology: Imaging vs. Imagining Multiresolution Deformations Multiresolution Deformations 4-D CT Example 4-D CT Example Exhale State Exhale State Inhale State Validation of Registration Validation of Registration 4-D CT Example Study A Point 1 2 3 4 5 6 Description 2nd branch of bronchial tree 3rd branch of bronchial tree 4th branch of bronchial tree Vessel bifurcation 1 Vessel bifurcation 2 Vessel bifurcation 3 X -5.37 -5.73 -6.50 -8.12 -8.06 -10.69 Exhale Y 0.98 2.12 2.77 3.37 -1.95 2.47 X -4.62 -5.40 -6.24 A A -8.12 -7.67 -10.78 Inhale Y -0.22 0.74 0.80 1.40 -3.20 1.16 Z -2.92 -5.92 -9.42 -11.42 -3.92 1.08 * ∆X -0.09 0.05 -0.03 B B -0.07 0.40 -0.07 Exhale' - Inhale ∆Y -0.25 -0.16 -0.11 -0.49 0.37 -0.29 ∆Z -0.44 0.09 -0.09 -0.18 0.29 0.16 σ 0.19 0.29 0.26 * Z -3.42 -5.42 -8.42 A A-9.92 A A -4.42 0.58 (x , y , z ) Inhale State deformed Study B all values in cm. Exhale' ( w/ TPS alignment ) -4.71 -0.47 -3.36 -5.35 0.58 -5.83 -6.27 0.69 -9.51 B B B B -8.19 0.91 -11.60 -7.27 -2.83 -3.63 -10.85 0.87 1.24 (x , y , z ) Exhale State Marc L Kessler, PhD Inhale State 28 Multimodality Imaging in Radiation Oncology: Imaging vs. Imagining Balter/ Balter/ UM UM 4-D Deformable Phantom Multiresolution Deformations 4-D CT Example Exhale State Inhale State deformed Ruan Ruan // UM UM We Are Not Really Splines ! Two Part Cost Function • intensity similarity measure • tissue-dependent deformation regularization No “stiffness” information Extracted Ribcage T * = arg min ESSD(T ) + α Evol(T ) T Exhale Deform Inhale Marc L Kessler, PhD 29 Multimodality Imaging in Radiation Oncology: Imaging vs. Imagining Two Part Cost Function Regularization Term Similarity term: Sum squared differences ESSD = ( f(x) - g(T.x) )2 dx Regularization term: Volume preservation Evol = wc(x) |det JT(x) – 1|2 dx Ruan Ruan // UM UM “Stiffness” Function Using “Prior” Information wc(x) using “stiffness” information Extracted Ribcage Marc L Kessler, PhD Exhale Deform Inhale 30 Multimodality Imaging in Radiation Oncology: Imaging vs. Imagining Balter Balter // UM UM Tissue Sliding Deal with different organs individually? Mean Square Difference Segmentation + Registration No No masking masking Mask Mask using using segmented segmented lung/abdomen lung/abdomen Note ribs (and (and tissue tissue )) near near diaphragm diaphragm Note ribs Ribs by lung lung registration registration Ribs not not affected affected by Product Comparison Mean Mean Square Square Difference Difference The The mean mean square square difference, difference, MSD, MSD, is is aa simple simple method method to to evaluate evaluate the the similarity similarity between between two two images, images, A A and and B, B, as as is is shown shown in in equation equation 7. 7. MSD MSD is is aa direct direct approach approach that that can can only only be be used used when when the the intensities intensities between between images images A A and and B B are are the the same, same, but but provides provides aa simple simple minimization minimization problem problem for for optimizing optimizing 13,14 registration. registration.13,14 MSD( A, B) = A− B www.ITNonline.net 2 Marc L Kessler, PhD 31 Multimodality Imaging in Radiation Oncology: Imaging vs. Imagining Don’t try this at home! Practical Aspects … do whatever you can to reduce the number of degrees of freedom of the image registration problem ! Using too many degrees of freedom will … increase computation time increase the likelihood of local minima likely decrease the overall accuracy www.itk.org Degrees of Freedom PET/CT MR - CT Degrees of Freedom 4D CT PET CT PET / CT Hybrid GE Discovery LS None ? Few Many T = Identity ? Marc L Kessler, PhD XPET = XCT ? 32 Multimodality Imaging in Radiation Oncology: Imaging vs. Imagining Not Always None! PET XPET = XCT JNM JNM 46:1488-96 46:1488-96 2005 2005 Not Always None! CT Not Always None! Not Always None! User Beware! MR - CT phantom in head frame CT artifact from respiration “burned in” to attenuation corrected PET Mechanically attach a coordinate system Marc L Kessler, PhD You still need to be careful ! 33 Multimodality Imaging in Radiation Oncology: Imaging vs. Imagining The Mechanics! Automated Registration FDG PET 11C Original PET C PET Original 11 CT Image Registration ) H(IPET PET 1.83 Data Fusion PET X-ray CT Synthetic Synthetic MR-PET MR-PET Image Image Original Original MR MR H(ICT) 3.98 Registered Registered MR MR DICOM 3 Parts 3 & 17 Manipulate CT to match FDG-PET - rotate / translate - DICOM 3 Parts 3 & 17 From From DICOM DICOM 3.3-2007, 3.3-2007, page page 209 209 A.39.2.1 A.39.2.1 Deformable Deformable Spatial Spatial Registration Registration IOD IOD Description Description The The Deformable Deformable Spatial Spatial Registration Registration Information Information Object Object Definition Definition (IOD) (IOD) describes describes spatial spatial relationships relationships between between images images in in one one or or more more frames frames of of reference reference via via deformation grids and transformation matrices. The deformations and deformation grids and transformation matrices. The deformations and transformations transformations describe describe to to an an application application how how to to sample sample data data from from one one or or more more Source Source RCSs RCSs into into the the Registered Registered RCS. RCS. The The Registered Registered RCS RCS is is the the Frame Frame of of Reference Reference specified specified within within an an instance instance of of this this IOD. Source RCS RCS Frame Frame of of IOD. The The IOD IOD may may specify specify that that only only aa subset subset of of the the entire entire Source Reference is affected by the transformation, by specifying specific frames of Reference is affected by the transformation, by specifying specific frames of image image SOP Source Frame Frame of of Reference. Reference. SOP Instances Instances that that use use the the Source The The deformation deformation is is described described as as aa grid grid of of offset offset vectors. vectors. Each Each grid grid element element contains contains 3 3 values values representing representing offset offset distances distances in in the the X, X, Y, Y, and and Z Z directions directions at at the the center center position position of of the the deformation deformation grid grid element. element. The The relationship relationship between between the the data data being being deformed deformed and and the the deformation deformation grid grid is is purely purely spatial. spatial. Therefore Therefore the the resolution resolution of of the the grid grid is is independent independent of of the the data data being being deformed. deformed. DICOM handles only up to affine ... (and most Tx planning systems) Marc L Kessler, PhD 34 Multimodality Imaging in Radiation Oncology: Imaging vs. Imagining New DICOM Objects ? Prospective OK Marc PET / CT Hybrid GE Discovery LS F = Identity Not Always Perfect! XB = XA User Beware! User Beware! CT artifact from respiration “burned in” to attenuation corrected PET Marc L Kessler, PhD 35 Multimodality Imaging in Radiation Oncology: Imaging vs. Imagining Why did that work? User Beware MR - CT phantom … in attach a head frame ouch … reproduce patient orientation as closely as possible using an immobilization device coordinate system to the patient! … Stereotactic Radiosurgery Why did that work? … reproduce patient orientation as closely as possible using an immobilization device Virtual CT device device Virtual PET PET -- CT Custom-molded Styrofoam cradle Thorax Board Sinmed BV Full 3D / 4D Deformation Warp Space / … Drag Objects Warp Objects / … Drag Space 18FDG is not very specific - it lights up … 18 a lot of tissues to some extent - it contains anatomic information … breath hold exhale CT similar to free breathing PET fairly Brock Brock // PMH PMH Parametric Marc L Kessler, PhD Freeform 36 Multimodality Imaging in Radiation Oncology: Imaging vs. Imagining How ? Prospective Prospective reproduce imaging geometry exactly attach coordinate system to patient • frames / fiducials Retrospective PET / CT Hybrid GE Discovery LS patient intrinsic • anatomy / shape / image intensities F = Identity Prospective ouch XB = XA Retrospective … attach a coordinate system to the patient! How do we determine T ? Interactive Tools … let the experienced user drive Automated Tools … let the computer drive … Stereotactic Radiosurgery Marc L Kessler, PhD 37 Multimodality Imaging in Radiation Oncology: Imaging vs. Imagining Interactive Registration Multi-resolution B-Splines Should we really describe this? Translate Yes, … go to next slide Rotate ? Deform No, … skip past this section Provide tools to transform and visualize! B-Spline Transformation Model B-Spline Transformation Model Transformation is built up using a set of weighted basis splines Transformation is built up using a set of weighted basis splines ∆X basis spline ∆X basis splines w w11 knot k11 w w11 X knots … these are “just like” the beamlets we use in IMRT Marc L Kessler, PhD k11 w w22 k22 X … these are “just like” the beamlets we use in IMRT 38 Multimodality Imaging in Radiation Oncology: Imaging vs. Imagining B-Spline Transformation Model B-Spline Transformation Model Transformation is built up using a set of weighted basis splines Transformation is built up using a set of weighted basis splines ∆X basis splines w w11 k11 knots ∆X w w22 k22 basis splines w w33 w w11 k33 X knots k11 w w22 k22 w w33 k33 w w44 k44 X … these are “just like” the beamlets we use in IMRT … these are “just like” the beamlets we use in IMRT B-Spline Transformation Model B-Spline Transformation Model Transformation is built up using a set of weighted basis splines Transformation is built up using a set of weighted basis splines ∆X weighted sum ∆X w w11 knots k11 w w22 k22 w w33 k33 w w11 w w44 k44 X β(X-kii) X’ = X + ∆X = X + Σ wiiWβ weights knots k11 w w22 k22 w w33 k33 w w44 k44 X β(X-kii) X’ = X + ∆X = X + Σ wiiWβ splines Marc L Kessler, PhD weights splines 39 Multimodality Imaging in Radiation Oncology: Imaging vs. Imagining B-Spline Transformation Model B-Spline Transformation Model Transformation is built up using a set of weighted basis splines Transformation is built up using a set of weighted basis splines ∆X w w11 knots k11 w w33 ∆X w w44 w w11 w w22 k22 k33 k44 X knots β(X-kii) X’ = X + ∆X = X + Σ wiiWβ weights k11 w w33 w w44 w w22 k22 k33 k44 X β(X-kii) X’ = X + ∆X = X + Σ wiiWβ splines weights splines B-Spline Transformation Model B-Spline Transformation Model Transformation is built up using a set of weighted basis splines Tighter knot spacing allows more local deformation ∆X w w11 knots k11 w w33 ∆X w w44 w w22 k22 k33 k44 X knots k11 k22 k33 k44 k55 k66 k77 k88 k99 k10 k11 10 11 X β(X-kii) X’ = X + ∆X = X + Σ wiiWβ … linear with respect to weights! Marc L Kessler, PhD 40 Multimodality Imaging in Radiation Oncology: Imaging vs. Imagining B-Spline Transformation Model B-Spline Transformation Model Tighter knot spacing allows more local deformation Tighter knot spacing allows more local deformation ∆X ∆X w w77 knots k11 k22 k33 k44 k55 k66 k77 k88 w w77 k99 k10 k11 10 11 X knots k11 k22 k33 k44 k55 k66 k77 k88 k99 k10 k11 10 11 X B-Spline Transformation Model B-Spline Transformation Model Tighter knot spacing allows more local deformation Tighter knot spacing allows more local deformation ∆X knots ∆X w w77 k11 k22 k33 k44 k55 k66 k77 k88 k99 k10 k11 10 11 X knots k11 k22 k33 k44 k55 k66 k77 k88 k99 k10 k11 10 11 X However, more degrees of freedom can lead to local minima Marc L Kessler, PhD 41 Multimodality Imaging in Radiation Oncology: Imaging vs. Imagining Regional Registration Can you tell what is different in the 2 images? Regional Registration Bones aligned, prostate region not aligned No Cropping The “answer” depends on the region defined! Regional Registration Dawson Dawson // PMH PMH What About Using Just a Few? Bones ignored, prostate region aligned Most of the motion of the liver seems to be rigid or affine! … some deformation does occur though. Cropping The “answer” depends on the region defined! Marc L Kessler, PhD 42 Multimodality Imaging in Radiation Oncology: Imaging vs. Imagining What About Using Just a Few? What About Using Just a Few? maybe ignore over a limited field-of-view maybe ignore over a limited field-of-view ? particularly poor MR CT ( Diagnostic ) ( Therapy ) anatomic-based data cropping CT registered to MR split-screen display What About Using Just a Few? What About Using Just a Few? maybe ignore over a limited field-of-view maybe ignore over a limited field-of-view anatomic-based data cropping anatomic-based data cropping CT registered to MR image-switch display Marc L Kessler, PhD CT registered to MR image-switch display 43 Multimodality Imaging in Radiation Oncology: Imaging vs. Imagining Limited Field-of-View Limited Field-of-View Rigid assumption used for regional registration Pick Your Battles Wisely! Marc L Kessler, PhD Oops! 44