Nuclear Medicine Physics Single Photon Emission Computed Tomography (SPECT) Jerry Allison, Ph.D. Department of Radiology Medical College of Georgia

Medical College of Georgia And Sameer Tipnis, Ph.D. A note of thanks to Z. J. Cao, Ph.D. Medical College of Georgia And Sameer Tipnis, Ph.D. G. Donald Frey, Ph.D. Medical University of South Carolina for Sharing nuclear medicine presentation content

Tomographic NM imaging (SPECT) Single Photon Emission Computed Tomography 2015 Nuclear Medicine Physics for Radiology Residents Sameer Tipnis, PhD, DABR

Scintillation Camera © Physics in Nuclear Medicine: Cherry, Sorenson and Phelps, 4th edition, 2012

Scintillation Camera PMT Event Positioning Network © Physics in Nuclear Medicine: Cherry, Sorenson and Phelps, 4th edition, 2012

Tomographic imaging (SPECT) Produce tomographic images by acquiring conventional gamma camera projection data at several angles around the patient Similar to CT

SPECT Provide 3-D images to eliminate overlaying and underlying activity of a slice better contrast more accurate lesion localization more demanding technically and longer data acquisition more severe image noise

SPECT data acquisition Generally, two detectors mounted at 180 or 90 on a rotation gantry

SPECT data acquisition a sequence of 2-D static images at different angular positions (views) detector rotation range 180º with 2 perpendicular detectors or 360º with 2 opposite detectors 45º RAO In principle, rotation of the gamma camera about 180° allows for the acquisition of sufficient projections for tomographic reconstruction. However, in practice, opposing views acquired 180° apart differ due to various factors (photon attenuation, depth dependent collimator response) and SPECT data are commonly acquired over 360°. 45º LPO

SPECT data acquisition Circular or elliptical orbit, which is better? closer to the patient  better spatial resolution

SPECT Image Acquisition (Improves spatial resolution) 2015 Nuclear Medicine Physics for Radiology Residents Sameer Tipnis, PhD, DABR

Data Collection: Configuration Non Cardiac Cardiac Delete??????

SPECT Image Acquisition Typically 2 camera heads, rotating around patient Projection images every 3 – 6 degrees ~ 30 s / projection, ~ 15 minutes total Matrix - 64 x 64 or 128 x 128 2015 Nuclear Medicine Physics for Radiology Residents Sameer Tipnis, PhD, DABR

Data Collection: Angular Stops 3 to 6 degrees is common a lesser number causes streaking a larger number does not improve image quality Step-shoot or continuous acquisition, which is better?????? Step & Shoot Characteristics Some loss of time Less Blur

View number for 360º SPECT number of views = matrix size 128 views

Each projection has 128 data points An image with 128 x 128 matrix Contains 128 projections Each projection has 128 data points Equivalent to 128 slice CT (i.e., 128 tomographic slices per rotation) 2015 Nuclear Medicine Physics for Radiology Residents Sameer Tipnis, PhD, DABR

Sinogram (for one of many slices) There are many (e.g. 64 or 128) sinograms per rotation. © Physics in Nuclear Medicine: Cherry, Sorenson and Phelps

© Physics in Nuclear Medicine: Cherry, Sorenson and Phelps

Back Projection Leads to blurring in image (streaks and star-like artifacts) Deleted Kuhl’s Brain Imaging

Filtered Backprojection Suppress blurring through filtering the projections A high-pass filter (ramp filter) can be used to suppress blurring A problem that is immediately apparent is the blurring (star-like artifacts) that occur. One would expect that a high-pass filter could be used to eliminate blurring, and that is the case. The optimal way to eliminate these patterns in the noiseless case is through a ramp filter. The combination of back projection and ramp filtering is known as filtered back projection.

Filtered Back Projection (of noiseless data) © Physics in Nuclear Medicine: Cherry, Sorenson and Phelps © Physics in Nuclear Medicine: Cherry, Sorenson and Phelps

Filter Amplitude Frequency Ramp filter Amplitude Ramp w/ rolloff filter Frequency Define K(L) & L In the absence of noise, a ramp filter works well For noisy images, a ramp filter with roll-off is required

Filter Ramp Ramp with some roll-off filter Suppresses blurring but enhances noise Ramp with some roll-off filter Smoothes the image and suppresses noise Trade off noise vs resolution Roll-off filter characteristics are adjustable

Filter Types © Physics in Nuclear Medicine: Cherry, Sorenson and Phelps

Filter Types (A) Butterworth: (less noise, more smooth) (B) Butterworth: (more noise, less smooth) 2015 Nuclear Medicine Physics for Radiology Residents Sameer Tipnis, PhD, DABR

Filter Applied filter is the product of: Ramp User selected/characterized filter Shepp-Logan Hahn Butterworth Weiner Hamming Hanning (MCG: Philips, generally turned off)

Filter Applied filter is the product of: Procedure for applying filter Ramp User selected/characterized filter Shepp-Logan (cut-off freq) Hann (cut-off freq) Butterworth (order (slope), critical freq (0.707 response) Procedure for applying filter 1D projection of each view is converted to spatial frequency using a Fourier transform Ramp filter with roll-off is applied in spatial frequency space (k-space) Filtered projection is recovered with inverse Fourier transform Back projections performed to reconstruct image

Selection of Filters for SPECT Filters trade noise for resolution No standard way to optimize filter choice Patient to patient variation Physician preferences Vendor recommendation Embellish from Cherry.

Iterative Reconstruction (IR) Filtered Back Projection has some limits Various corrections needed Attenuation Compton Scatter Ordered Subsets Expectation Maximization (OSEM) is a common iterative reconstruction algorithm Improve comments on limitations of filtered back projection.

Calculation Includes: Attenuation Scatter Blur with depth Assume Some Image (I) Calculate Projections (P’) Calculation Includes: Attenuation Scatter Blur with depth Compare to Measured Projection (P) Use P’ & P to form corrections Image estimate may be a uniform image Assumed image undergoes forward projection to produce sinogram Form New Image (I’) Is I-I’< * Done

Iterative Reconstruction Slow compared to filtered back projection Commonly used for PET Being used increasingly in SPECT IR used for ~all Philips SPECT at MCG Siemens C-Cam?????? Has become the standard for SPECT?????

Image recon - Iterative Common IR recon is the OSEM For OSEM, # iterations (I) and # subsets (S) affect image quality  # (I/S)   noise, but sharper images 2015 Nuclear Medicine Physics for Radiology Residents Sameer Tipnis, PhD, DABR

Iteration 1 2015 Nuclear Medicine Physics for Radiology Residents Sameer Tipnis, PhD, DABR

Iteration 5 2015 Nuclear Medicine Physics for Radiology Residents Sameer Tipnis, PhD, DABR

Iteration 10 2015 Nuclear Medicine Physics for Radiology Residents Sameer Tipnis, PhD, DABR

Brain Phantom IR OSEM FBP 2015 Nuclear Medicine Physics for Radiology Residents Sameer Tipnis, PhD, DABR

Non-filter Noise Factors Collimator Matrix 64 x 64 128 x128 Slice thickness Time per stop/ Number of Stops Administered Dose Deleted: Number of matrix element rows

Data Collection Whole Image is collected for each view 64 x 64 or 128 x 128 Each row makes a slice Multiple slices can be added to reduce noise Anything higher than 128 x 128?

Data Collection: Counts Determination of Number of Image Counts Activity in patient Time per stop Number of Stops

Attenuation Correction Like all radionuclide imaging there is a problem due to attenuation. Correction can be important for judging the activity of lesions

Attenuation correction s traveling smaller paths through pt (nearer to camera) have less attenuation compared to those from deeper in pt  for AC can be assumed or measured Chang (assumed), Measured - Gd rods (older) or CT (new) CT can be non-diagnostic (low power, cone-beam) or fully diagnostic depending on the scanner model 2015 Nuclear Medicine Physics for Radiology Residents Sameer Tipnis, PhD, DABR

Attenuation in SPECT I0 = I1e+a (attenuation corrected intensity) Probability of detection / correct intensity I0, dependent on the depth at which  originates  need to know “a” 2015 Nuclear Medicine Physics for Radiology Residents Sameer Tipnis, PhD, DABR

Chang’s AC method Image first reconstructed without AC Contours of image used to estimate t for each projection,  assumed to be constant Average ACF determined for each pixel (x,y) from all projections Reconstructed image corrected pixel-by-pixel Works well for area with approximately constant attenuation like head, abdomen but not for areas like chest / thorax 2015 Nuclear Medicine Physics for Radiology Residents Sameer Tipnis, PhD, DABR

I(x) = I0e-mx (x is thickness of tissue between pixel & detector) Chang’s Method Assume uniform attenuation m linear attenuation coefficient m is ~0.15/cm for Tc-99m for soft tissue I(x) = I0e-mx (x is thickness of tissue between pixel & detector) 0.15/cm is the narrowbeam attenuation coefficient for 140 keV photons in soft tissue Use of the broadbeam coefficient, 0.12/cm, helps to account for attenuation and scatter in uniform radioactivity distributions In Chang’s method, m is often set to ~0.12/cm to better account for Compton scatter

Uniform phantom with evenly distributed 99mTc Chang method Proper AC Low counts in center 2015 Nuclear Medicine Physics for Radiology Residents Sameer Tipnis, PhD, DABR

SPECT/CT

AC in SPECT/CT Accurate / realistic -map obtained for each projection using CT  values used in Chang’s algorithm to correct pixel-by-pixel AC here is more realistic (since  is not assumed to be constant) Current SPECT/CT systems use this method 2015 Nuclear Medicine Physics for Radiology Residents Sameer Tipnis, PhD, DABR

Philips Astonish NM Recon Software The Astonish: Ordered Subsets Expectation Maximization (OSEM) Compensation for the blurring effects of the collimator built into the reconstruction Resolution Recovery allows recovery of some of the original resolution Astonish uses the distance from the detector to the object of interest recorded as a function of angle by the camera during acquisition and geometric properties of the specific collimator Has become the standard for SPECT?????

Philips Astonish NM Recon Software Used on essentially all Philips NM SPECT images at MCG Iterations: 4 Subsets: 16 CT data used for attenuation correction (except brains due to EEG electrode artifacts) Chang’s AC invoked for brains Has become the standard for SPECT?????

Common SPECT Problems Patient motion System Alignment (Center of rotation issues) Collimator issues Distance issues Loss of resolution with distance

Center of Rotation SPECT assumes heads always look at a constant central rotation point

COR (Spatial Alignment) Image will have blurring and circular artifacts COR must be tested periodically for all heads SPECT of three point sources Generally done with system QC software SPECT of 3 point sources ?????

COR correction SPECT image of a uniform phantom on a camera with poor COR correction Incorrect COR correction introducing a ring artifact, degrading spatial resolution 2015 Nuclear Medicine Physics for Radiology Residents Sameer Tipnis, PhD, DABR

COR correction COR correction is used when reconstructing tomo-graphic data to correct for minor misalignment between the center of the image and the axis of rotation. COR corrections are stored in a correction table and are applied automatically after a data set has been acquired. 2015 Nuclear Medicine Physics for Radiology Residents Sameer Tipnis, PhD, DABR

Collimator Issues Collimators are not completely uniform A high count flood must be stored to correct for collimator non-uniformities 20 M for 5% for 128 x128 Check or delete “20 M for 5% for 128 x128”

Patient Studies Advantages Disadvantages No overlapping structures 3 dimensional lesion locations Fusion with high resolution images (CT, MRI) Disadvantages Time consuming (motion) Images are noisy