SPECT RECONSTRUCTION FOR DUMMIES :SPECT RECONSTRUCTION FOR DUMMIES


OBJECTIVES :OBJECTIVES REVIEW THE BASIS OF SPECT IDENTIFY DIFFERENT DOMAINS AND EXPLAIN THE TRANSFORMATION DEFINE RECONSTRUCTION DISCUSS THE METHODS OF RECONTRUCTION DISCUSS FILTERS USED TO ENHANCE IMAGES


WHAT IS SPECT ? :WHAT IS SPECT ? Single Photon Emission Computed Tomography Acquisition mode that takes many 2D planar images and reconstructs them into 3D Produces a higher contrast, higher resolution image of patients anatomy than planar acquisition


DOMAINS :DOMAINS Spatial domain- represents data in familiar terms such as counts per pixel in the x and y coordinates Frequency domain – object information is represented as a sum of sinusodial waves represented by a frequency spectrum


RECONSTRUCTION DOMAINS :RECONSTRUCTION DOMAINS Reconstruction process is done to retrieve the spatial distribution of a radiotracer from the projection Images can be reconstructed in two domains spatial or frequency Reconstruction algorithms fall into to categories: analytic (FBP) and iterative


SPECT RECONSTRUCTION :SPECT RECONSTRUCTION Two ways to reconstruct an image: filtered back projection iterative reconstruction (FBP) can be used in spatial domain or frequency domain (IR) uses computer estimates of transaxial slices and compares it to the original projection


Signal vs. Noise :Signal vs. Noise Signal is considered the part of data gathered that produced the image Noise is extra data that does not contribute to the final image Noise is caused by: photon scattering statistical variations random electronic fluctuations


BACK PROJECTION :BACK PROJECTION FBP is used in spatial and frequency domain The combination of back projection and ramp filtering is known as filtered back projection. In spatial domain uses the nine point smooth technique to improve images Improvements in frequency domain are done through the use of filters


BACK PROJECTION :BACK PROJECTION Data is used to create multiple transaxial images Projections are ran back through the image along the same lines from where the photon was emitted from Area where lines from different angles intersect represent STAR ARTIFACT


Projection Views :Projection Views


Projection Views :Projection Views


Projection Views :Projection Views 0  0  0  0  30  48  46  26  0  0  0  0  0


9 POINT SMOOTHING :9 POINT SMOOTHING Smoothing- Partially redistributes the counts from the pixels with the highest pixel to its immediate neighbors Technique - the counts from the adjacent pixel and central pixel are averaged - central pixel is then replaced by the average - process is repeated pixel to pixel


FILTERING :FILTERING Mathematical technique applied during reconstruction to improve the appearance of the image Removes star artifact Removes noise


TRANSFORMATION OF DOMAINS :TRANSFORMATION OF DOMAINS Any image pattern can be represented by a combo of sine and cosine waves in cycles per pixel Transformation was developed by French mathematician Fourier


FREQUENCY DOMAIN :FREQUENCY DOMAIN Frequency spectrum – amplitude is plotted in the y axis , frequency is plotted in the x axis Nyquist frequency- highest frequency useful for showing that two adjacent points are separate Cut off frequency – highest frequency used by a filter Frequencies higher than Nyquist cannot contribute to the final image


FILTERS :FILTERS High pass filters- cut out low frequencies resulting in sharp corners and fine detail ex. Ramp filter Low pass filters- cuts off high frequencies which gives us overall shape of the image ex. Butterworth, Hanning, Hamm, Parzen, Weiner


IMAGES :IMAGES Low frequencies yield the overall shape of the image High frequencies yield the sharp corners and fine detail of the image


FILTERS :FILTERS Too little filtering of high frequencies result in excessive grainy texture Too much filtering of high frequencies results in over smoothing of an image (loss of edge definition)


ITERATIVE METHOD :ITERATIVE METHOD Uses computer estimates to reconstruct images Compares estimated images to original images Short cut methods of iterative: MLEM-maximum likelihood expectation maximization OSEM-ordered subsets expectation maximization


ITERATIVE METHOD :ITERATIVE METHOD Estimation process continues until difference between estimate and actual projection view is less than a certain threshold Incorporates correction factors for attenuation and scatter Produces significantly less star artifact than FBP Slowly replacing filtered back projection


SUMMARY :SUMMARY SPECT- Single Photon Emission Computed Tomography Spatial and Frequency domains are used for image reconstruction Reconstruction is done to separate signal from noise and enhance images Two reconstruction methods used are filtered back projection and iterative reconstruction Some low pass filters used are Butterworth, Hamm, Hanning, Weiner and Parzen A high pass filter is the ramp filter


SUMMARY :SUMMARY Iterative reconstruction produces less star artifact than filtered back projection Low frequencies yield the overall shape of the image High frequencies yield the sharp corners and fine detail of the image


References :References Christian, P. Nuclear Medicine And PET/CT Mosby 6th edition. 2007. pgs 279-286. Zeissman, H. Nuclear Medicine: The Requisites of Radiology. Mosby 3rd edition. 2006. pgs 51-62. Powsner, R. Nuclear Medicine Physics. Blackwell 2nd edition. 2006. pgs. 85-110 Images retrieved on www.google.com.


Questions :Questions What are the two domains used in filtered back projection? Spatial and Frequency Domains


Slide 26:True or False?? A signal is the information that produces the image and noise is the extra data that does not contribute to the image. TRUE


Slide 27:3. Which type of reconstruction produces less star artifact: Iterative or Filtered Back Projection


Slide 28:4. A Butterworth filter is an example of: High pass filter Low pass filter Medium pass filter Spatial filter


Slide 29:5. True or False Nyquist frequency is the highest frequency useful for showing that two adjacent points are separate. TRUE