MR Data for Treatment Planning and Stereotactic Procedures: Sources of Distortion, Protocol Optimization, and Assessment (Preview of TG117 Report) Debra H. Brinkmann Mayo Clinic, Rochester MN Acknowledgements TG-117 Use of MRI Data in Treatment Planning and Stereotactic Procedures and Quality Control Procedures Deb Brinkmann, chair Kiaran McGee R. Jason Stafford Ed Jackson Steve Goetsch Outline MR in Treatment Planning Spatial accuracy of MR images Impact of Scanner (strength / configuration) Impact of Pulse Sequence / Parameters Distortion assessment / QC MR in Tx Planning Overview MR in RTP Advances in planning and delivery necessitate improved target delineation MR soft tissue contrast Overview MR in RTP MR - Functional / biological information Further improve target definition / extent Target severity with dose Khoo, 2006 BJR Chang, 2006 MedPhys 1
Overview MR in RTP Image registration To correlate MR-delineated structures to CT Kessler, 2006 BJR 52 cm MR narrow 70 cm MR wide Overview MR in RTP Spatial Distortions Can be 1cm Scanner-Dependent Distortions External Magnetic Field Inhomogeneity, Environmental Gradients Nonlinearities, Scale Factor Errors, Eddy Currents RF Slice Profile Patient/Object-Induced Distortions Chemical Shift Magnetic Susceptibility 80 85 cm CTsim MR basics MR signals depend on Larmor equation: ω = γb 0 Localization depends on linear relationship b/w space & resonance frequency Established by linear gradients ω = γ ( B 0 + G r) MR basics In general: RF pulse is applied together with a gradient to generate a signal from a particular sub-volume Signals are spatially encoded using gradients for image localization Imaginary Complex MR signals are detected Magnitude φ Magnitude and phase components Real Reconstruction to create an MR image MR basics Perturbed MR signal (magnitude, phase) affects: Presence of artifacts Image quality Interference with relationship between space and resonance frequency: x = ω/γ -B Leads to geometric distortions 0 G Shifting x Warping Intensity gradients Spatial Accuracy 2
Scanner-dependent distortions Result of: Design compromises Imperfections Drifting / failure of specific components Constant Across multiple imaging sessions Allows for regular testing External magnetic field inhomogeneities: Desire high uniformity over entire volume For linear relationship b/w space & frequency Perfect uniformity not technically feasible Imperfections change resonant frequency Result spatial misregistration Shimming to improve B0 homogeneity: Passive shims (pieces of metal in bore) Installed during initial calibration Active shims (superconducting shim circuits) Adjusted at calibration, preventative maintenance Resistive shims (linear, non-linear shim circuits) Adjusted during imaging Shim imaging volume vs. whole field Non-linear efficacy - application dependent External magnetic field inhomogeneities: Homogeneity maintained over finite spherical volume Vendor specification: DSV (diameter of spherical volume) Imaging outside of DSV subject to distortions Shift depends on strength of applied gradient x = ω/γ - B 0 G x Environmental Magnetism: Electromagnetic shielding used in site design Stray magnetism can affect MR acquisition Example: Garbage truck near scanner Iron in truck magnetized Affected B0 homogeneity Result: severe distortion With Garbage Truck Without Garbage Truck Environmental Magnetism: Importance of QC: More subtle effects might not be apparent QC procedures needed to catch such errors Example detected impact from steel beams for construction placed near MR suite Courtesy of Kiaran McGee 3
Gradient Nonlinearities: Spatial encoding achieved with gradients Linearly mapping position with frequency Deviations from linearity due to: deviations in rise time peak amplitude physical design Deviations with distance from isocenter primary source of distortion over majority of imaging volume Baldwin, Med Phys 2007 Gradient Nonlinearities: Conventional 2D sequences: pin-cushion or barrel effect in-plane potato chip effect on image plane bow-tie effect on slice thickness Sumanaweera, Neurosurg 1994 Gradient Nonlinearities Example: warping in-plane Baldwin, Med Phys 2007 Example: warping along slice select direction Gradient Nonlinearity Corrections: Vendors provide in-plane corrections Assume distortion to gradient amplitude is constant distortion with phantom Apply to reconstructed images after patient data acquired Some but not all vendors provide corrections along slice encoding axis Wang, Med Phys 2004 Gradient Field Nonlinearity Effects - In-Plane Distortion Slice Plane at Isocenter 20 cm FOV White circles: with gradient nonlinearity correction Black circles: without gradient non-linearity correction Maximum error w/o correction: ~ 4.5 mm at ±10 cm from isocenter Maximum error w/ correction: < 1 mm at ±10 cm from isocenter Gradient Field Nonlinearity Effects - In-Plane Distortion Slice Plane at 20 cm from Isocenter 20 cm FOV White circles: with gradient nonlinearity correction Black circles: without gradient non-linearity correction Maximum error w/o correction: ~ 5.5 cm at ±10 cm from iso Maximum error w/ correction: < 2 mm at ±10 cm from iso MDACC MR Research Courtesy of Ed Jackson, MDACC MDACC MR Research Courtesy of Ed Jackson, MDACC 4
Gradient Field Nonlinearity Effects - Slice Distortion Large gradient coil Slice Profile Phantom Small gradient coil Frequency Ultra-wide short bore 490 Hz/pixel no vendor corrections (a) 490 Hz/pixel 2D vendor corrections (b) 125 Hz/pixel 2D vendor corrections (c) Effects are exacerbated with ultrashort bore magnets! MDACC MR Research Courtesy of Ed Jackson, MDACC Phase gradient corrections do not correct for distortions due to resonance offsets Courtesy of R. Jason Stafford, MDACC Gradient Nonlinearities Example: pelvis with vs. without corrections CT MR, no corrections MR, GDC corrections MR, GDC + additional point-by-point corrections Chen, Med Phys 2006 Reported distortion values Wang et al, MRI 2004 (five 1.5T systems) 5-26mm uncorrected 240x240x240mm 3 3-12mm 2D vendor corrections applied 240x240x240mm 3 <2.5mm 2D vendor corrections applied 95mm sphere Baldwin et al, Med Phys 2004 (3T system) <7mm uncorrected 260x260x200mm 3 Doran et al, PMB 2005 (1.5T system) 25mm uncorrected 320x200x340mm 3 Chen et al, Med Phys 2006 (0.23T system) <5mm 2D vendor corrections applied 360x360x200mm 3 <5mm 2D vendor corrections applied 480x480x200mm 3 Eddy Currents: Generated in conducting materials Metal dewer,, gradient coils, RF coils exposed to time varying magnetic field Faraday s s law of induction Changing gradient fields Creates perturbing magnetic field Lenz s s law Result spatial distortion Vendors provide some correction method RF non-uniformity uniformity: RF energy generates a detectable signal Delivered as a pulse (RF pulse) Design criteria for RF waveform Exact shape variable May not be designed to maintain uniform signal Issue for advanced techniques Gradient linearity also impacts slice profile Evaluate RF profile to verify width 5
Patient/object-induced induced distortions Result of: Composition of patient or object Unique: Can vary dramatically b/w patients Must be considered for each imaging situation Chemical shift (of the first kind): Produces shift in resonant frequency E.g. 220 Hz decrease for 1.5 T (ω 0 = 63.8 MHz) Misregistration when BW/ pixel < chemical shift Manifest along frequency encoding direction Produces banding Banding only lipid signal shifted for voxels with mixed composition Chemical Shift of the first kind 16 Hz 31 Hz 61 Hz Magnetic Susceptibility: Relates net magnetization to the applied magnetic field Materials in magnetic field become magnetized Creates changes in magnetic field at interfaces Complex, depending on many factors: Susceptibility difference across interface Shape and orientation of the interface with B0 Strength and polarity of gradients 122 Hz 244 Hz 488 Hz Magnetic Susceptibility: Perturbations produce distortion, signal loss Magnetic Susceptibility Relatively small for soft tissues Undetectable for many applications Difference b/w tissue & air: ~ 9 x 10-6 * distortions Air-tissue interfaces for air cavities External surface of patient 16 Hz 31 Hz 61 Hz *Schenk, Med Phys 1996 122 Hz 244 Hz 488 Hz 6
Magnetic Susceptibility: Difference between metal and tissue is large E.g., titanium ~ 20 times larger than soft tissue* Impact of Scanner Courtesy of Kiaran McGee *Schenk, Med Phys 1996 Impact of Scanner Design includes compromises & trade-offs No single design for all performance spec s Systems optimized to meet subset of applications MR data for use in RTP Each scanner used should be characterized The physicist needs Understanding of distortion sources Ability to quantitate image distortion Impact of Scanner Field strength options: 0.2-1.0 T ( Open,, permanent or resistive magnet) 1.5, 3.0 T (Cylindrical bore, superconducting magnet) LOW field strength advantages: patient-induced induced distortions artifacts from metal objects (e.g. brachytherapy) HIGH field strength advantages: signal-to to-noise (image quality) resolution for metabolites (MRS) Impact of Scanner Impact of Scanner OPEN magnet: Advantages Flexibility in patient positioning Increased patient access Drawbacks Significant external field inhomogeneities scanner-dependent distortions www.gehealthcare.com GE 0.7T OpenSpeed www.medgear.org Phillips 1.0T Panorama MR simulator NARROW, CYLINDRICAL LONG bore: High performance systems (e.g. cardiac) High resolution imaging (e.g. CNS) Advantages Increased field homogeneity scanner-dependent distortions Drawbacks Narrow bore patient comfort 7
Impact of Scanner WIDE, SHORT bore: Advantages Increased magnet aperture Treatment position claustrophobia Drawbacks www.medical.siemens.com Siemens 1.5T Espree Wide / short bore Sacrifice performance, field homogeneity, field strength Decreased homogeneity increased distortion Impact of Sequence Impact of Sequence Sensitivity to distortion: Gradient echo sequences Most sensitive to distortion sources Inhomogeneity effects accumulate throughout acquisition Conventional spin echo sequences 180 refocusing pulse reduces distortion Fast spin echo sequences Least sensitive to off-resonance effects Multiple 180 refocusing pulses and short TEs 3D vs. 2D sequences: For standard rectilinear imaging, spatial distortion due to resonance offsets: Manifest along the frequency encoding axis Phase encoding direction not affected 3D acquisitions use phase encoding along slice encoding direction Less distortion for 3D vs. equivalent 2D acquisitions Impact of Sequence Advanced acquisition techniques: Majority rely on echo planar imaging or EPI* EPI techniques collect a train of echoes Uninterrupted accumulation of phase Very sensitive to field inhomogeneities and eddy current effects PE: severe shifting or compression of objects FE: shearing of object Object induced inhomogeneities considerable local distortions distortions with increasing field strength Bandwidth per pixel: BW minimizes resonance offsets Field inhomogeneity,, chemical shift, magnetic susceptibility If Δf f > BW/pixel, shift will result Magnitude depends on pixel dimensions Trade-off: BW will SNR SNR (voxel vol. Ny NEX / BW) *Mansfield, Br J Radiol 1978 16 Hz 488 Hz 8
BW / pixel for Patient-Induced Distortions: Distortions with increasing field strength Distortions with decreasing BW/pixel Δf f = γ B 0 ppm, (3.5ppm for CS, 9.0ppm* for MS) # pixels = Δf f / (BW/pixel) (depends on pixel dimensions) # pixels 0-5 -10-15 Chemical Shift Artifact 0.2 T 1.5 T 3.0 T # pixels (+/-) *Schenk, Med Phys 1996 50 40 30 20 10 Maximum Susceptibility Artifact 0.2 T 1.5 T 3.0 T Spatial resolution: Magnitude depends on pixel dimensions Typical pixel resolution.75 1.5 mm/pixel Higher resolution reduces physical dimension of shift FOV will resolution Trade-off: resolution will SNR SNR (voxel vol. Ny NEX / BW) -20 0 50 100 150 200 250 300 Bandwidth (Hz/pixel) 0 0 50 100 150 200 250 300 Bandwidth (Hz/pixel) Scenario # Reference 1 2 3 4 5 6 7 8 B 0 1.5 1.5 1.5 1.5 3.0 3.0 3.0 3.0 0.5 delta_fat_water 215 215 215 215 429 429 429 429 72 FOV 250 250 220 320 250 250 320 320 260 BW (total FOV) 32000 64000 32000 64000 32000 64000 96000 84000 20000 Num Freq Enc Points 256 256 256 320 256 256 320 320 256 Num Phase Steps 256 256 256 320 256 256 320 320 256 BW per pixel 125 250 125 200 125 250 300 263 78 NEX 1 1 2 1 1 1 0.5 0.5 4 chemical shift 1.68 0.84 1.48 1.07 3.35 1.68 1.43 1.63 0.93 susceptibility shift 2.26 1.13 1.99 1.44 4.51 2.26 1.93 2.20 1.25 relative SNR 100 71 110 93 200 141 107 114 91.2 relative acquisition time 100 100 200 125 100 100 62.5 62.5 400 resolution (frequency) 0.98 0.98 0.86 1.00 0.98 0.98 1.00 1.00 1.02 resolution (phase) 0.98 0.98 0.86 1.00 0.98 0.98 1.00 1.00 1.02 T Hz mm Hz Hz mm mm % % mm mm Lipid Suppression: Lipid signal can be nulled If not clinically relevant To eliminate chemical shift effects Both kinds of chemical shift artifacts Several imaging techniques Spectral saturation Inversion recovery Dixon technique Lipid Suppression: Spectrally selective saturation pulses Uses RF pulse to saturate spins precessing at resonant freq- uency of fat Affected by poor shimming: Incomplete fat saturation Inadvertent suppression of water Courtesy of Kiaran McGee Impact of Sequence Lipid Suppression: Inversion recovery techniques (STIR) Dixon technique Requires multiple data acquisitions (two TE) Lipid & tissue signals in and 180 out of phase Add / subtract for lipid- and tissue-only images J Ma, JMRI 2006 IP Water-only T1, FSE STIR OP Fat-only 9
Frequency Encoding Direction: Resonance offsets manifest along frequency encoding direction Can manipulate to visualize such distortions (in-plane) Repeating scan with reversed gradient Repeating scan swapping frequency & phase Cost additional scan Distortion Assessment / QC Current Guidance AAPM and ACR have published acceptance test and QC documents Do not address necessary QC program and image acquisition optimization goals when MRI data used for procedures in which spatial accuracy is critical ACR Weekly QC protocol Geometric accuracy criteria 2mm over 148mm x 190mm Courtesy of Kiaran McGee Works in Progress TG-117: Use of MRI Data in Treatment Planning and Stereotactic Procedures and QC Procedures Review physical bases for spatial accuracy limitations in MRI Provide guidance with examples for reducing or eliminating the effects of distortion Propose QC tests for systems used for applications requiring high spatial accuracy Assessment/QC for new application Initiate Scanner/ Upgrade Yes Modify/ Upgrade or Scanner? No No Review feasibility of pursuing application Report Findings to Supervising Physician Performance Adequate for? Yes Establish Routine Distortion QC 10
Step 1: Identify Volumetric coverage (FOV, cranio-caudal caudal extent) Spatial resolution (voxel dimensions) Spatial accuracy (tolerance, volume) Bore diameter, RF coils (tx position) MR compatibility of immobilization, applicators Pulse sequence(s) (contrast) Step 2: Equipment and Scanner capabilities Identify magnet, gradients, coils, sequences needed to achieve required performance specifications Existing scanner? Upgrades needed? New system? Phantoms for geometric distortion assessment Acceptance / commissioning tests Routine QC Analysis tools (IT support for programming, networking) Step 3: Baselines (Acceptance / Commissioning) Characterize scanner subsystems External magnetic field (homogeneity) Gradients (linearity, correction algorithms, eddy current compensation ) RF (slice profile) Step 3: Existing tests (ACR, AAPM) Purpose: maintain diagnostic image quality Do not assess spatial fidelity over RTP volumes Can be used with modifications Assess over the desired imaging volume Using application specific imaging parameters Step 3: Specific Protocol Optimization 2D vs. 3D FOV, in-plane resolution Gradient echo vs. Spin echo Signal suppression techniques requirements identified Initiate Scanner/ Upgrade System characterized Yes Modify/ Upgrade Over volume of interest or Scanner? Review feasibility of pursuing application No Step 4: Performance Adequate? Report Findings to If not: Supervising Physician Modify application? Upgrade scanner? Initiate modification/upgrade & re-evaluate evaluate Or report unacceptable accuracy No Performance Adequate for? 11
Step 5: Establish QC program Measure scanner dependent distortions Phantom of known geometry Verify constancy of spatial fidelity Drifting, failure Establish Routine Distortion QC Environmental magnetism Over the volume of interest Yes Performance Adequate for? Establish Routine Distortion QC Establishing a QC program Distortion phantom for routine QC Define the QC phantom requirements Volume to test over Fiducial spacing Performance Adequate for? Fiducial resolution (in-plane 5 pixels) Dimension compatibility (RF coil, stereotactic frames, immobilization) Identify / modify / develop application-specific QC phantom Yes Establishing a QC program QC testing Establish imaging protocol Determine testing frequency Develop analysis tools Automated evaluation Automated reporting Establish Routine Distortion QC Establish procedures when QC fails Yes Performance Adequate for? Summary Spatial fidelity in MR images Source of geometric distortions Impact of scanner characterization Impact of image acquisition parameters Vendor supplied correction methods Importance of assessing distortions Over the volume of interest With the same parameters to be used clinically Importance of appropriate MR QC program 12