# Filtering in SPECT Image Reconstruction.

**ABSTRACT** Single photon emission computed tomography (SPECT) imaging is widely implemented in nuclear medicine as its clinical role in the diagnosis and management of several diseases is, many times, very helpful (e.g., myocardium perfusion imaging). The quality of SPECT images are degraded by several factors such as noise because of the limited number of counts, attenuation, or scatter of photons. Image filtering is necessary to compensate these effects and, therefore, to improve image quality. The goal of filtering in tomographic images is to suppress statistical noise and simultaneously to preserve spatial resolution and contrast. The aim of this work is to describe the most widely used filters in SPECT applications and how these affect the image quality. The choice of the filter type, the cut-off frequency and the order is a major problem in clinical routine. In many clinical cases, information for specific parameters is not provided, and findings cannot be extrapolated to other similar SPECT imaging applications. A literature review for the determination of the mostly used filters in cardiac, brain, bone, liver, kidneys, and thyroid applications is also presented. As resulting from the overview, no filter is perfect, and the selection of the proper filters, most of the times, is done empirically. The standardization of image-processing results may limit the filter types for each SPECT examination to certain few filters and some of their parameters. Standardization, also, helps in reducing image processing time, as the filters and their parameters must be standardised before being put to clinical use. Commercial reconstruction software selections lead to comparable results interdepartmentally. The manufacturers normally supply default filters/parameters, but these may not be relevant in various clinical situations. After proper standardisation, it is possible to use many suitable filters or one optimal filter.

**2**Bookmarks

**·**

**339**Views

- [Show abstract] [Hide abstract]

**ABSTRACT:**Renal cortical scintigraphy with technetium-99m (Tc-99m) dimercaptosuccinic acid (DMSA) is the method of choice to detect acute pyelonephritis and cortical scarring. Different acquisition methods have been used: Planar parallel-hole or pinhole collimation and single photon emission tomography (SPECT). This study compared planar parallel-hole cortical scintigraphy and dual-head SPECT for detection of cortical defects. We retrospectively reviewed 190 consecutive patients with 380 kidneys and 200 DMSA scans referred to rule out renal cortical scarring. The diagnoses were 52 vesicoureteric reflux, 61 recurrent urinary tract infection, 39 hydronephrosis, 20 renal impairment, and 18 hypertension. All patients were imaged 3 h after injection of Tc-99m DMSA with SPECT and planar imaging (posterior, anterior, left, and right posterior oblique views). For each patient, planar and SPECT images were evaluated at different sittings, in random order. Each kidney was divided into three cortical segments (upper, middle and lower) and was scored as normal or reduced uptake. The linear correlation coefficient for the number of abnormal segments detected between planner and SPECT techniques was calculated. From 200 DMSA scans, 100 scans were positive for scar in SPECT images, from which only 95 scans were positive for scar in planner imaging. Out of the five mismatched scans, three scans were for patients with renal impairment and high background activity and two scans were for very small scars. No significant difference was seen in the average number of abnormal segments detected by planar versus SPECT imaging (P = 0.31). The average correlation coefficient between was high (r = 0.91 - 0.92). Tc-99m DMSA renal cortical scanning using SPECT offers no statistically significant diagnostic advantage over multiple views planar imaging for detection of cortical defect.Indian journal of nuclear medicine : IJNM : the official journal of the Society of Nuclear Medicine, India. 01/2015; 30(1):26-30. - SourceAvailable from: Robin de Nijs[Show abstract] [Hide abstract]

**ABSTRACT:**Due to low counts in an 111In single photon emission computed tomography (SPECT) scan, a large part of the head was missing in the reconstructed images on Philips Extended Brilliance Workspace (EBW) and IntelliSpace Portal (ISP) workstations. This problem occurred for the ordered subsets expectation maximization (OSEM) algorithm with and without resolution recovery (Astonish), but not for filtered backprojection (FBP) or maximum likelihood expectation maximization (MLEM). There were also underflow problems because the images are stored as integers resulting in a loss of intensity resolution and color banding. Philips EBW2.0 and ISP5.02 workstations upscale low-count images, but the result is not always optimal, for example, in the case of low counts in one part and more counts in another part of an image. On these workstations, the missing head artefact problem could be resolved by applying a Hann pre-filter (with a cutoff at the Nyquist frequency, which only influences the filtering) in the reconstruction process. Upscaling of the projection data prior to reconstruction did not recover the head in the images, neither did limiting the reconstructed volume to the low-count part of interest. Underflow problems were partially solved by the new version 2.0 of the Philips EBW and ISP stations, although situations could arise where underflow still poses a problem. A solution for the underflow problems is to upscale the raw projection data before reconstruction. While this results in a pure upscaling of the FBP reconstruction, the effect in iterative statistical reconstruction is not only upscaling of the intensities because the assumption of Poisson statistics of the data is violated. However, the influence of this last matter seems limited. Reconstruction of studies with low counts in relevant areas should be performed with care. Reconstruction artefacts and scaling issues can easily arise.EJNMMI Physics. 09/2014; 1(10). - SourceAvailable from: Pedro Fragoso CostaA Technologist´s Guide - Myocardial Perfusion Imaging (Revised), 1 edited by Helen Ryder, Giorgio Testanera, Vanessa Veloso Jerónimo, Borut Vidovic, 10/2014: chapter Chapter 8 - Image Processing and Software: pages 32; EANM., ISBN: 978-3-902785-09-1

Page 1

Hindawi Publishing Corporation

International Journal of Biomedical Imaging

Volume 2011, Article ID 693795, 14 pages

doi:10.1155/2011/693795

Review Article

FilteringinSPECT Image Reconstruction

Maria Lyra andAgapi Ploussi

Department of Radiology, Radiation Physics Unit, University of Athens, 76, Vas. Sophias Ave., Athens 11528, Greece

Correspondence should be addressed to Maria Lyra, mlyra@med.uoa.gr

Received 25 January 2011; Accepted 5 April 2011

Academic Editor: M’hamed Bentourkia

Copyright © 2011 M. Lyra and A. Ploussi. This is an open access article distributed under the Creative Commons Attribution

License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly

cited.

Single photon emission computed tomography (SPECT) imaging is widely implemented in nuclear medicine as its clinical role in

the diagnosis and management of several diseases is, many times, very helpful (e.g., myocardium perfusion imaging). The quality

of SPECT images are degraded by several factors such as noise because of the limited number of counts, attenuation, or scatter

of photons. Image filtering is necessary to compensate these effects and, therefore, to improve image quality. The goal of filtering

in tomographic images is to suppress statistical noise and simultaneously to preserve spatial resolution and contrast. The aim of

this work is to describe the most widely used filters in SPECT applications and how these affect the image quality. The choice

of the filter type, the cut-off frequency and the order is a major problem in clinical routine. In many clinical cases, information

for specific parameters is not provided, and findings cannot be extrapolated to other similar SPECT imaging applications. A

literature review for the determination of the mostly used filters in cardiac, brain, bone, liver, kidneys, and thyroid applications

is also presented. As resulting from the overview, no filter is perfect, and the selection of the proper filters, most of the times,

is done empirically. The standardization of image-processing results may limit the filter types for each SPECT examination to

certain few filters and some of their parameters. Standardization, also, helps in reducing image processing time, as the filters

and their parameters must be standardised before being put to clinical use. Commercial reconstruction software selections lead

to comparable results interdepartmentally. The manufacturers normally supply default filters/parameters, but these may not be

relevant in various clinical situations. After proper standardisation, it is possible to use many suitable filters or one optimal filter.

1.Introduction

Tomography is a noninvasive imaging technique that is used

to generate cross-sectionals images of a three dimensional

(3D) object without superimposing tissues. Tomography

can be categorized in transmission tomography such as

computed tomography (CT) and emission tomography like

single photon emission computed tomography (SPECT) and

positron emission tomography (PET). Computed tomog-

raphy is a technique based on X-ray transmission through

a patient to create images of sections (slices) of the

body. Photon emission computed tomography and positron

emission tomography provide 3D image information about

the radionuclide injected into the patient that shows the

metabolic and physiological activities within an organ.

In tomographic techniques, projections are acquired

from many different angles around the body by one or more

rotating detectors. These data are then reconstructedand put

together to form 3D images of the body. The reconstruction

of tomographic images is made by two methods: filtered

backprojection and iterative methods [1].

The quality of the final tomographic image is limited

by several factors. Some of these are the attenuation and

scatter of gamma ray photons, the detection efficiency and

the spatial resolution of the collimator-detector system [2].

These factors have as a result poor spatial resolution, low

contrast, and high noise levels. Image filtering techniques

are very important in tomography as they strongly affect the

quality of the image.

Image filtering is the term used for any operation that is

applied to pixels in an image. It is a mathematical process

by which images are suppressed in noise and also includes

smoothing, edge enhancement and resolution recovery.

Filters are used during reconstruction and applied to

data in frequency domain. The goal of the filtering is to

compensate for loss of detail in an image while reducing

Page 2

2International Journal of Biomedical Imaging

noise. The application of filters is the most common method

to reduce high-frequency noise component in projection

images. In this way, filters can greatly improve the image

resolution and limit the degradation of the image. There are

severaltypesoffiltersusedinmedicalimagingandthechoice

of the appropriate filter is a headache in clinical practice [3].

The aim of this article is to describe the most commonly

used filters in SPECT imaging by analytical techniques.

These filters are applied in filtered back projection (FBP)

reconstruction techniques. Filtering can also be considered

as a postprocessing step in iterative reconstruction. Though

many times iteratively reconstructed images need to be

postfiltered, as they tend to be noisy, special dedicated

iterative filters are not established yet to be included in

commercial software. We present also choices of filters for

some SPECT examinations that are common in clinical

routine, as they are suggested in the literature.

2.SPECTImage Acquisition

Nowadays single-photon emission computed Tomography

(SPECT) is widely used in nuclear medical imaging. SPECT

is a nuclear medical tomographic imaging technique that

represents the distribution of an administered radioactive

tracerwithinanorgan.Theradiopharmaceuticalemitssingle

gammarayphotons.SPECTsystemsuseoneormoregamma

camera mounted on a gantry so that the detector can rotate

around the patient. From the acquired one dimensional

projection data from different views around the object, two

dimensional (2D) planar projections images are obtained in

many evenly spaced angles around the patient and provide

an estimate of 3D distribution of the radiotracer using image

reconstruction from multiple projections. Some systems

acquire the images during their rotational movement, while

others stop and record (stop and shoot) an image at selected

angles. In SPECT the projections images are generally

acquired over a full 360-degree or 180-degree arc (in case

of SPECT myocardium perfusion study or kidneys SPECT

acquisition), on a matrix of 64 ∗ 64 or 128 ∗ 128 pixels.

Typically the projections are acquired every 3–6 degrees

and the total scan time is about 15–20 minutes. The 2D

projection-imagesarefirstcorrectedfornonuniformitiesand

then mathematical algorithms are used to reconstruct 3D

matrices of selected planes from the 2D projection data.

3.SPECTImage Reconstruction

The purpose of reconstruction algorithms is to calculate

an accurate 3D radioactivity distribution from the acquired

projections. There are two methods to reconstruct SPECT

images, either iteratively or by FBP technique.

3.1. Iterative Reconstruction Method. Iterative reconstruction

starts with an initial estimate of the image [4]. Most of

the times the initial estimate is very simple, for example a

uniform activity distribution. Then a set of projection data is

estimatedfromtheinitialestimateusingamathematicalpro-

cess called forward projection. The resulting projections are

compared with the recorded projections and the differences

between the two are used to update the estimated image. The

iterative process is repeated until the differences between the

calculated and measured data are smaller than a specified

preselected value. The iterative reconstruction methods

include algebraic methods like the algebraic reconstruction

technique (ART) and statistical algorithms like maximum

likelihood expectation maximization (MLEM) or ordered-

subsets expectation maximization (OSEM) [1].

3.2. Filtered Backprojection Method (FBP). FBP is an analyti-

calmethodthatisstillthemostwidelyusedinclinicalSPECT

becauseofitssimplicity,speed,andcomputationalefficiency.

It consists of two steps: filtering of data and back projection

of the filtered data [5].

In 2D acquisition, each row of projections represents

the sum of all counts along a straight line through the

depth of the object being imaged. Back projection technique

redistributes the number of counts at each particular point

back along a line from which they were originally detected.

This process is repeated for all pixels and all angles. The

limited number of projection sets has as a result the creation

of a star artifact and the blurring of the image. To eliminate

this problem the projections are filtered before being back

projected onto the image matrix. It has to be noticed that the

backprojection process is taken place in spatial domain while

data filtration is done in the frequency domain.

3.3. Image SPECT Filtering. The image restoration process

is an example of Fourier spectrum filtering. Once a Fourier

Spectrum has been generated for an image, it can be filtered

so that certain spatial frequencies can be modified, enhanced

or suppressed. This filtered spectrum can then be inverse

transformed to generate a filtered image with, for example,

sharpened or smoothed features. A feature we need to

consider in more detail is the spatial frequency nature of

the image data itself. Images are generally sampled digitally

using a square matrix composed of pixels, the size of which

dictates how well a digital image approximates its analogue

counterpart.

The filters used in FBP are simply mathematical equa-

tions that vary with frequency. The filters used in SPECT

imaging can vary to achieve different purposes such as star

artifact reduction, noise suppression, or signal enhancement

and restoration.

The choice of filter for a given image reconstruction

task is generally a compromise between the extent of noise

reduction and fine detail suppression (and of contrast

enhancement in some cases) as well as the spatial frequency

pattern of the image data of interest.

3.3.1. Filtering to Reduce the Star Artifact

Ramp Filter. The ramp filter is a high pass filter that does

not permit low frequencies that cause blurring to appear in

theimage.Infrequencydomain,itsmathematicalfunctionis

given by (1).

?

where kx, kyare the spatial frequencies.

HR

kx,ky

?

= k =

?

k2

x+k2

y

?1/2, (1)

Page 3

International Journal of Biomedical Imaging3

Point source

(a) (b)(c)

Figure 1: A simple representation of filtered back projection. (a) Acquisition of three projections. (b) Backprojected projections. (c) Filtered

backprojected projections.

The Ramp is a compensatory filter as it eliminates the

star artifact resulting from simple backprojection. Because

the blurring is only appeared in the transaxial plane, the

filter, is only applied in that plane [5]. The filter as shown in

Figure 1(a)), is linearly proportional to the spatial frequency.

High pass filters sharpen the edges of the image (areas in

an image where the signal changes rapidly) and enhance

object edge information. A severe disadvantage of high pass

filtering is the amplification of statistical noise present in the

measured counts. In order to reduce the amplification of

high-frequencies the ramp filter is always combined with a

low-pass filter.

3.3.2. Filtering to Reduce Noise. The common method to

reduce or remove statistical noise in a SPECT image is

the application of smoothing filters. These filters are low-

pass filters which allow the low frequencies to be retained

unaltered and block the high frequencies. Low-pass filters

are characterized mainly by two parameters—the “cut-off

frequency” and the “Order” (or the “Power”). The cut-off

frequency (or roll-off frequency) defines the frequency above

which the noise is eliminated. The filter function is defined

to be zero for all frequencies above cut-off frequency. The

Nyquist (Nq) frequency—the highest frequency that can be

displayed in an image—is apparently the highest cut-off

frequency for a filter. The cut-off frequency is expressed in

cycles per pixel or as a fraction of the Nq frequency. The

cut-off frequency varies typically from 0.2 to 1.0 times the

Nq frequency. The value of the cut-off frequency determines

how the filter will affect both image noise and resolution. A

highcut-offfrequencywillimprovethespatialresolutionand

therefore much detail can be seen but the image will remain

noisy. A low cut-off frequency will increase smoothing but

will degrade image contrast in the final reconstruction.

The parameter Order controls the slope of the filter

functionandcharacterizesthesteepnessoftherolloff.Ahigh

order will result in a sharp fall. Sometimes, the term power

instead of order is used. The power is twice the order.

There is a number of low-pass filters that are available

for SPECT reconstruction. The most commonly used are

discussed below.

Butterworth Filter. Butterworth filter is the more usual

choice in nuclear medicine. The butterworth filter is a low-

pass filter. It is characterized by two parameters: the critical

frequency which is the point at which the filter starts its roll

off to zero and the order or power [6]. As it is mentioned

earlier, the order changes the slope of the filter. Because of

this ability of changing not only the critical frequency but

also the steepness of the roll-off, the butterworth filter can

do both, smoothes noise and preserves the image resolution.

A butterworth filter in spatial domain is described by:

B?f?=

1

?

1+?f/fc

?2n?, (2)

where f is the spatial frequency domain, fc the critical

frequency and n the Order of the filter (Figure 3).

A ramp function and a butterworth function of variable

order and cut-off (critical) frequency, are multiplied to form

the fourier filter used in the FBP process (Figure 4).

Hanning Filter. The Hanning filter is a relatively simple low-

pass filter which is described by one parameter, the cut-off

(critical) frequency (Figure 5) [7].

The Hanning filter is defined in the frequency domain as

follows:

⎧

⎪⎪⎩

where f are the spatial frequencies of the image and fmthe

cut-off (critical) frequency.

In signal processing, the Hann window is a window

function, called the Hann function, named after Julius

Ferdinand von Hann, an Austrian meteorologist. The use of

the Hann window is called “Hanning”, as a signal to apply

the Hann window to a digital signal processing. http://en

.wikipedia.org/wiki/Hann function.

The Hanning (Hann) filter is very effective in reducing

image noise as it reaches zero very quickly; however, it does

not preserve edges (Figure 5).

H?f?=

⎪⎪⎨

0.50 +0.50 cos

?π f

fm

?

,0 ≤

otherwise,

??f??≤ fm,

0,

(3)

Page 4

4International Journal of Biomedical Imaging

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

0

0.2

0.4

0.6

0.8

1

0 0.2 0.40.60.81

1.2

02468

−0.2

−0.4

x

f(x)

HR

k

(a)

(b)

Figure 2: The Ramp filter: (a) Ramp filter in frequency domain. (b) Ramp filter in spatial domain [3].

0

00.1 0.3 0.5

0.2

0.2

0.4

0.4

0.6

0.8

1

1.2

A

(A) n = 2, fc= 0.1

(B) n = 8

(C) n = 32

BCD

(D) n = 2, fc= 0.3

(E) n = 8

(F) n = 32

EF

Spatial frequency (cycle/pixel)

Relative magnitude

Figure 3: Butterworth smoothing filter six curves by different

fc and n parameters (equation (2)). A, B, C curves created by

critical frequency fc = 0.1c/pixel and order n equal to 2, 8, 32

correspondingly. D, E, F curves created by critical frequency fc =

0.3c/pixel and order n equal to 2, 8, 32 similarly [6].

Hamming Filter. The Hamming filter is also a low pass

filter, which presents a high degree of smoothing, named

afterRichardWesleyHamming,anAmericanmathematician

famousincomputerscience.AstheHanningfilter,ithasonly

0

0

20 406080 100

0.2

0.4

0.6

0.8

1

Amplitude

Butterworth

Multiplied

50% level

Ramp

Nyquist frequency (%)

Cutoff: 40%

Order: 3

Figure 4: Illustration of the Butterworth filtering process. A Ramp

function and a Butterworth function (of Order 3 and cut-off

frequency 40% of Nq frequency) are multiplied to form the Fourier

filter used in the FBP process. Generated by Kieran Maher, 2006,

accessed in http://en.wikibooks.org/wiki/File:NM16 14.gif.

asingleparametertodescribeitsshape,thecut-offfrequency.

The mathematical definition is shown as (4) [7].

H?f?=

⎧

⎪⎪⎩

⎪⎪⎨

0.54 +0.46 cos

?π f

fm

?

,0 ≤

otherwise,

??f??≤ fm,

0,

(4)

where f are the spatial frequencies of the image and fmthe

cut-off frequency.

As it can be observed the only difference with the

Hanning filter is on the amplitude at the cut-off frequency.

Page 5

International Journal of Biomedical Imaging5

012

(cycles/cm)

0

0.5

1

Hanning filter

Ramp filter

Amplitude

Cut-off frequency 0.8c/cm

Figure 5: Hanning filter and Ramp in FBP reconstruction.

Parzen Filter. The Parzen filter is another example of low

pass filter and it is defined in the frequency domain as [7],

??f??−6??f??

???f??

fm

?

?2

×

?

1 −

??f??

fm

?fm

???f??≥ fm

? ???f??<fm

??f??< fm

2

?

,

P?f?=

⎧

⎪⎪⎩

⎪⎪⎨

2??f??

0,

1 −

??f??

fm

?3

,

2

<

?

,

?,

(5)

where f are the spatial frequencies of the image and fmthe

cut-off frequency.The Parzen filter is the most smoothing

filter; it eliminates high-frequency noise, but it also degrades

the image resolution [3].

Shepp-Logan Filter. The Shepp-Logan is one more filter that

belongs to the family of low pass filters. Its mathematical

equation is shown as (6) [8].

S?f?=

2fm

?π?sin??f??π/2fm

??.

(6)

TheShepp-Loganfilterproducestheleastsmoothingandhas

the highest resolution.

Numerous types of filters exist, and all filters aim, except

for the restoration filters, at reducing frequency information

through an amplitude-adjusting function between 0 and

1Nq. The interpretation and comparison of SPECT studies

is beclouded by the use of too many different filter types.

Optimal parameters have been calculated [3] for But-

terworth or Hanning filters to match the shape of various

existing filter types. Butterworth filters cannot approximate

any other kind of filter shape since the amplification

of the high-frequency components always asymptotically

approaches zero, whereas for the Hann filter, high-frequency

components can be set to zero. This is demonstrated for

0

0.2

0.4

0.6

0.8

1

H

00.51

k

Filters functions’ curves

Butterworth

Parzen

Hann

Shepp-Logan

Figure 6: Shepp-Logan, Butterworth, Hann, Parzen filter func-

tions’. (from Van Laere et al., (2001), modified ) [3].

the approximation of a Hann filter by Butterworth matching

(Figure 6). A Shepp-Logan filter can be very accurately

matched to a Butterworth filter with the appropriate param-

eters. A Parzen filter is closely matched by a Hann filter with

cut-off 1 (Figure 6).

From the practical point of view, all filter shapes

can be fairly accurately addressed by a specific cut-

off/order/restoration combination of Butterworth and Hann

filtering.

3.3.3. Filtering to Enhance the Signal. A low-pass filter may

smooth image to a high degree that does not permit discern-

ing small lesions, leading to contrast loss. For this reason a

thirdclassoffilters,calledenhancementorrestorationfilters,

is used in SPECT imaging. The restoration filters enhance

the signal with a simultaneous reduction of noise without

resolution lost. Metz and Wiener are two types of resolution

recovery filters that have been used in nuclear medicine

image processing.

Metz Filter. Metz filter is a function of modulation transfer

function (MTF), and it is based on the measured MTF of the

gamma camera system. The MTF describes how the system

handles or degrades the frequencies. The Metz restoration

filter is defined in the frequency domain as

M?f?= MTF?f?−1?

where f is the spatial domain and x is a parameter that

controls the extent to which the inverse filter is followed

before the low-pass filter rollsoff to zero [9]. Equation (7)

l −

?

l −MTF?f?2?x?

,(7)

Page 6

6International Journal of Biomedical Imaging

is the product of the inverse filter (first term) and a low pass

filter (second term).

The Metz filter is count dependent. Figure 7 shows the

Metz filter plotted for six different total image counts [10].

From Figure 7 results that, as the counts increase, more

resolution recovery occurs (filter rises farther above 1.0),

together with less suppression (filter moves farther to right)

[10].

Wiener. The Wiener filter is based on the signal-to-noise

ratio (SNR) of the specific image. The one dimensional

frequency domain form of the Wiener filter is defined as

W?f?= MTF−1×

MTF2

?

MTF2+N/O

?,(8)

where MTF is the modulation transfer function of the

imaging system, N is the noise power spectrum, and O is

the object power spectrum [11]. As with the Metz filter, the

Wiener is the product of the inverse filter (which shows the

resolution recovery) and the low pass filter (which shows the

noise suppression). In order to apply the Wiener filter it is

necessary to know a priori the MTF, the power spectrum of

the object and the power spectrum of the noise. It has to be

notedthatis,impossibletoknowexactlytheMTFortheSNR

in any image. As a result, the mathematical models used to

optimize both Metz and Wiener filters are uncertain [3].

3.4. Parameters Determining the Choice of the SPECT Filter

Type. Today,gammacamerasystemsofferachoiceofvarious

filters which may be selected depending on the type of

examination. The filter choice depends on [3, 12]:

(i) the energy of the isotope, the number of counts and

the activity administration.

(ii) the statistical noise and the background noise level.

(iii) the type of the organ being imaged.

(iv) the kind of information we want to obtain from the

images.

(v) the collimator that is used.

The choice of the filter must ensure the best compromise

between the noise reduction and the resolution in the image.

4.Typeof FiltersDependingon Typeof Study

The selection of the proper filter and the determination of

filter parameters is a major problem in clinical routine. In

this section, the filters used for widespread applications of

SPECTarelistedastheyaredescribedintheliterature.Image

filtering is an important, though mostly subjectively applied,

image-processing parameter. It is shown that ramp, Hann

and Butterworth filters are the most commonly used image

pre- and postprocessing filters. In many clinical evaluations,

literature does not provide useful information for specific

parameters of the imaging filters. In most clinical routine

cases the choice of a filter is done empirically, and the use

0

0

0.5

0.5

1

1.5

2

Filter

0.12

Spatial frequency (cycles/pixel)

0.250.37

Metz filter functions

Figure 7: Plot of Metz filter for total counts of 20.000, 50.000,

100.000, 200.000, 500.000, and 1 million counts from lowest to

highest curve [10].

(a)

(b)

(c)(d)

(e) (f)

Figure 8: Transverse slices of kidneys’. Various pre- postfiltered in

FBPreconstructionwereappliedwithdifferenteffectsontheimages

Filters used were (a) prefilter Hanning (cut-off 0.8cm−1), postfilter

Ramp. (b) prefilter Butterworth (cut-off 0.5cm−1power value 10),

postfilter Ramp. (c) prefilter Butterworth (cut-off 0.8cm−1, power

value 10), postfilter Ramp. (d) only Ramp prefilter applied—no

other smoothing filter. (e) prefilter Ramp, postfilter Hanning (cut

off 0.8cm−1) (f) prefilter Ramp, postfilter Butterworth (cut-off

0.8cm−1, power value 10). Study has been completed in Radiation

Physics Unit, Department of Radiology, University of Athens.

of limited filter types, in an attempt to standardise image-

processing approaches, may lead to better diagnostic com-

patibility and interpretation of interdepartmental results. In

Figure 8,theeffectofpre-orpostfilteringbyramp-Hanning-

Butterworth filters is shown, in coronal slices of a SPECT

renal study of a 6-month old boy.

Page 7

International Journal of Biomedical Imaging7

4.1. Cardiac SPECT. Cardiac SPECT has an important

clinical role in the detection of myocardial perfusion and

the diagnosis of ischemic heart disease. The commonly used

radiotracers for cardiac SPECT are Thallium-201 (201Tl)

and Technetium-99m (99mTc) labeled agents such as99mTc-

Sestamibi and99m-Tetrafosmin. In clinical practice, Hanning

filters were preferred for

for99mTc images [2]. In the literature, there are extensive

studies about the determination of the appropriate filter for

myocardial SPECT imaging.

Takavar et al. (2004) [13] studied the determination

of the optimum filter in99mTc myocardial SPECT using

a phantom that simulates the heart’s left ventricle. Filters

such as Parzen, Hanning, Hamming, and Butterworth and a

combination of their characteristic parameters were applied

on the phantom images. The cut-off frequency of 0.325Nq

and 0.5Nq gave the best overall result for Hanning and

Hamming filters, respectively. For Butterworth filter order

11 and cutoff 0.45Nq gave the best image quality and size

accuracy.

A determination of the appropriate filter for myocardial

SPECT was conducted by Salihin Yussoff and Zakaria [7].

The filters’ functions evaluated in this study included Butter-

worth, Hamming, Hanning, and Parzen filters. From these

filters, 272 combinations of filter parameters were selected

and applied to the projection data. The study suggested that

Butterworth filter succeeds the best compromise between

SNR and detail in the image while Parzen filter produced the

best accurate size.

The same group [14] investigated the relationship

between the optimum cut-off frequency for Butterworth

filter and lung-heart ratio in99mTc myocardial SPECT. A

linear relationship between cut-off frequency and lung-heart

ratio had been found which shows that the lung-heart ratio

must refer in each patient in order to choose the optimum

cut off frequency for Butterworth filter.

Links et al. (1990) [11] examined the affect of Wiener

filter in myocardial perfusion with201Tl SPECT. The study

wad done in 19 dogs and showed that Wiener filter

improves the quantization of regional myocardial perfusion

deficits.

In a201Tl gated SPECT study in patients with major

myocardial infraction [15], a Butterworth filter of order 5

with six cut-off frequencies (0.13, 0.15, 0.20, 0.25, 0.30,

0.35 cycle/pixel) were successively tested. The report showed

that filtering affect end-diastolic volume (EDV), end-systolic

volume (ESV) and left ventricular ejection fraction (LVEF).

Marie et al. (2005) [16] suggested that the best results for

cardiac gated SPECT image reconstruction with201Tl were

achieved using a Butterworth filter with an order of 5 and

cut-off frequency 0.30 cycles/pixel.

201Tl images and Butterworth

4.2. Brain SPECT. Brain SPECT is a powerful diagnostic

tool for evaluating neurologic and psychiatric diseases. Brain

SPECT provides a measure of cerebral blood flow (CBF)

and it is very useful for functional imaging of subcortical

structures of the brain. There are currently two commercial

radiotracers for brain SPECT imaging: Iodium-123 labeled

amphetamine (IMB) and99mTc hexamethylpropyleneamine

oxime (HMPAO). Due to the low SNR in this type of study

the choice of the optimum filter is difficult enough.

Groch and Erwin (2000) [5] showed that the most

suitable filter for

the Butterworth filter with order 10 and 0.5Nq cut-off

frequency. This filter gave the best compromise between

noise and spatial resolution with respect to Hamming filter.

In another report [17], the optimization of Butterworth

filter for brain SPECT imaging was studied. The aim of the

work was to find a relationship between the total counts and

the optimal cut off frequencies of the Butterworth filter. The

study proved that as the number of total counts increased the

optimal cut-off frequency linearly increased within a specific

range of counts.

Raeisi et al. (2007) [18] examined Ramp, Shepp-Logan,

Hanning, Hamming, Butterworth, Metz, and Wiener filters

in data from brain SPECT. The study suggested that both

Metz and Wiener give the maximum resolution and contrast

while Butterworth generate the best image quality.

99mTc-HMPAO brain SPECT study is

4.3. Other SPECT Studies. Although myocardial and brain

SPECT studies are the most widespread applications in

tomographic nuclear medicine examinations, there are

several other organs’ SPECT studies that were not very

commonlyused in clinicalroutine. Inthis time, SPECTdiag-

nosticandquantitativevalueisrecognisedascomplementary

assistance in the clinical diagnostic procedures, and accurate

volume estimations by SPECT are feasible when accurate

corrections are performed [19]. Some of them are bone,

liver, lungs, kidneys, and thyroid SPECT examinations. For

these applications, the most popular filters are Butterworth

and Hanning with different critical frequency values for

Hanning and various power and critical frequencies with

Butterworth filter. In many clinical cases, information for

specific parameters is not provided and filters’ parameters

findings cannot be evaluated and categorized per organ

study.

Bone. SPECT is an important diagnostic tool in nuclear

medicine for evaluating a detail image of the bones and

especially for detecting malignant. There are limited reports

in the literature for the appropriate filter in bone SPECT.

However,Butterworthfilterseemstoprovidemoreefficiently

anatomic details than other types of filters [20, 21]. Image-

dependent Metz filters have been shown to provide consis-

tently good image quality for bone study [22].

Liver. disease can be imaged using SPECT to determine

the existence of sarcoma, hepatic tumour, haemangiomas,

metastases, cyst, glycogen storage disease, and so forth, using

99mTc sulfur colloid (SC) [23]. In a SPECT study for the

anatomy of normal liver, Carrasquillo et al. (1983) [24]

suggested a modified Butterworth-ramp filter for the image

reconstruction. King et al. (1984) [10] showed that two

dimensional filtering, before and after reconstruction, using

the Metz and Wiener filters can improve significantly the

qualityofliverSPECTimages.Becauseofthehigh-countrate

and the high SNR in liver SPECT images, filters with a high

cut-off frequency are recommended to be used [3].

Page 8

8 International Journal of Biomedical Imaging

Transaxial slices

Liver

Spleen

Liver-spleen SPECT

(a)

3D liver

(b)

Figure 9: A liver-spleen study SPECT and 3D SPECT. (a) Transverse slices, prefiltered only by Ramp filter could emerge a small piece of the

rupturedspleen.(b)3DSPECTreconstructionbyramp-Hanningfilters(Hanningcut-offfrequencyequalto0.8cycles/cm),athreshold25%

and 15% gradient, could show the liver and a short fracture in the middle of right lobe of the liver but misses any residual spleen fragment.

(Study has been offered by Radiation Physics Unit, Department of Radiology, University of Athens).

Liver

Spleen

fragment

(a)

Liver haemangioma

R

(b)

Figure 10: 3D liver surface images. (a) Liver and spleen fragment caused by accident. Images reconstructed by FBP, prefiltered by Hanning

(criticalvalue0.8cm−1)andrampfilter.(b)Haemangiomaofliver.ImagesreconstructedbyFBP,prefilteredbyButterworth(cutoff0.5cm−1,

power factor = 10) and ramp filter. (Studies offered by Radiation Physics Unit, Department of Radiology, University of Athens).

Figure 9 shows a young boy’s liver-spleen study following

a car accident, searching for residual spleen pieces.

Two more cases of liver 3D SPECT images, reconstructed

by FBP, and different filters applied in (Figure 10).

Renal. SPECT by99mTc-DMSA is recommended to be used

instead of or complementarily to planar scintigraphy as

the preferable study to help especially in paediatrics with

early diagnosis, followup, and monitoring of the effects

of treatment in acute pyelonephritis and possible scars

formation [25–28].

In a renal SPECT study with99mTc-DMSA, De Sadeleer

et al. (1996) [25] suggested the use of a Butterworth filter

with an order of 7 and a cut-off frequency of 0.55Nq,

for the reconstruction of the projection data. According to

Groshar et al. (1997) [26], a Hanning filter with a cut-off

frequency of 0.5cycle/cm was applied in the data, in a kidney

SPECT imaging with99mTc-DMSA for best results.

In a study, by Yen et al. (1996) [27] for monitoring pae-

diatric acute pyelonephritis by99mTc-DMSA renal SPECT

imaging, a Metz prefilter was applied and transverse images

were reconstructed with back projection and a ramp filter to

show signs of acute pyelonephritis not indicated in planar

renal images.

A semiquantitative evaluation of cortical damage to the

kidneys, in children, was performed by tomographic renal

Tc99m-DMSAstudies.ReconstructionbyFBPusedHanning

filter (critical frequency 0.8cm−1and attenuation correction

0.12cm−1). The result of this procedure was the calculation

of three integrated over volume (IOV) indices that offer a

quantitative comparison of the planar, tomographic, and 3D

reconstructed images [28].

Recently,ina99mTc-DMSArenalcorticalSPECTimaging

study by dual head gamma camera, reconstruction was

performed similarly on both cameras using a Hann pre-

filter (cutoff frequency, 0.9cm−1; order, 0) and a Butter-

worth postfilter (cutoff frequency, 0.5cm−1; order, 10) with

two iterations and 10 subsets for the detection of renal

parenchyma focal defects [29].

Sheehy et al. (2009) have compared two methods

of reconstructing99mTc-dimercaptosuccinic acid (DMSA)

renal SPECT data—ordered subset expectation maximiza-

tion with OSEM-3D and FBP—in children in terms of

improving image quality and reducing the radiopharmaceu-

tical activity and radiation dose. Authors do not indicate the

Page 9

International Journal of Biomedical Imaging9

filters and relative parameters that were applied during FBP

[30]. OSEM-3D was described by R¨ omer et al. (2006) as an

iterative SPECT reconstruction algorithm that is performed

by using OSEM with 3-dimensional resolution recovery,

whichisappliedinthex, y,andz directions.Theyhadfound

that this approach, as compared with FBP, substantially

improves SPECT image quality and can be performed with

fewer gamma photon counts [31].

Lungs. SPECT techniques were, up to few years ago, used in

clinical diagnosis only by a limited number of centers. Given

the improvements in sensitivity and diagnostic accuracy

that has generally accompanied the transition from two-

dimensional planar to three-dimensional (3D) imaging,

SPECT technique in ventilation/perfusion (V/P) scintig-

raphy historically, one of the most commonly performed

diagnosticstudiesinnuclearmedicine,issuperiorincontrast

resolution and improved anatomical detail compared with

V/P perfusion scintigraphy, in the diagnosis of perfusion

embolism [32].

Gutte et al. (2010) concluded that V/Q SPECT should

be preferred in diagnosing of perfusion embolism. In their

study, SPECT datasets were attenuation corrected using the

low-dose CT acquisition with iterative reconstruction using

the software Autospect+ and Astonish with three iterations

and 16 subsets [33].

An automated linear registration algorithm based on

the maximization of mutual information was applied by

Reinartz et al. (2006) to the V/Q scans using the Hermes

Multimodality software including processing filters. This

automated analysis leads to a significant improvement in the

detection rate of pathologic lesions [34].

Harrisetal.(2007)showthatobjectiveanalysisofSPECT

V/Q scintigraphy provides a useful tool to help reduce the

number of nondiagnostic scintigraphy results. In their study

SPECT data were reconstructed to 128 slices using Ordered

Subset Expectation Maximization iterative reconstruction (8

iterations, 4 subsets), and smoothed with a postreconstruc-

tion three-dimensional Butterworth filter [35].

A comparison of usefulness of SPECT versus planar

lung scintigraphy in suspected pulmonary embolism, in

daily practice was completed by Weinmann et al. (2008).

Reconstruction of coronal, sagittal and transverse slices was

done by FBP followed by two iterations with a 5-order

Butterworth filter and a cut-off frequency at 0.45Nq [36].

An example of lungs’ perfusion embolism study by

SPECT is following (Figure 11).

A method for lungs’ volume determination by SPECT

and 3D SPECT images has been demonstrated [37]. Recon-

struction was performed by quantitative FBP by Hann filter

(critical frequency 0.9) and Chang attenuation correction

order 0, coefficient 0.11 in the GE Xeleris2 image processing

system. Phantom volume calculations were completed under

conditions similar to those of the patients’ studies. The

method assists to the accurate interpretation of perfu-

sion scans by volume, semi quantitative lung perfusion

index [LPI] and pulmonary improvement factor [PIF]

determination (Figure 12).

Lungs’ SPECT

RL

RL

LLLL

Coronal slices

Figure11:Twosequentialcoronalslicesindicatingmildpulmonary

embolism in left lung lobe (LL) not indicating in planar images.

Image reconstruction has been completed by FBP and Butterworth

filter (critical frequency 0.5, order 10). Courtesy of M. Gavrilelli

(MSc, “Medical Imaging Center” Athens, Greece).

Thyroid. SPECT volume estimation is an important tool for

dosimetry measurements and radionuclide therapy activity

dose determination.

Inacomparativestudyforthyroidvolumedetermination

by SPECT, Zaidi (1996) used the third order Butterworth

filterwithacut-offfrequencyequalto0.4Nq,forreconstruc-

tion of transaxial images of one pixel thickness. Images were

obtained without any correction and with two correction

methods. A slightly lower value of the attenuation coefficient

(μ = 0.12cm−1rather than μ = 0.15cm−1for99mTc) was

accepted better in quantifying thyroid volume by SPECT

[38].

ThyroidvolumeestimationswereperformedbyvanIsselt

et al., (2003), in patients with Graves’ disease [39]. The

planar images were subjected to filtering and thresholding,

and a standard surface formula was used to calculate the

thyroid volume. With SPECT, the iteratively reconstructed

thyroid images were filtered, and after applying a threshold

method, an automatic segmentation algorithm was used

for the volume determinations. Transmission scans, by two

gadolinium-153 (153Gd) line sources, were reconstructed

with FBP and were corrected for down scatter of99mTc into

the153Gd window. For the emission scan an iterative maxi-

mum likelihood reconstruction algorithm with attenuation

correction and window-based scatter correction as well as

resolutionrecoverywasused.Fornoisereduction,a3Dedge-

preserving 3×3×3-point median filter was applied.

Many times, phantoms of known dimensions have been

used in evaluating the accuracy of results in a methodology.

Bahk et al., (1998) used an acrylic thyroid phantom in their

study for pinhole SPECT imaging in normal and morbid

ankles. The phantom was subjected to planar, SPECT and

pinhole SPECT acquisitions. The gamma camera system

was connected to an Icon data processor that enabled

image reconstruction using the FBP algorithm and a Butter-

worth filter. The ankles’ SPECT scintigraphy was performed

immediately after pinhole scan by 360◦detector rotation.

The FBP algorithm and a Butterworth filter were used for

reconstruction as in the phantom study [40].

Page 10

10International Journal of Biomedical Imaging

R lobeL lobe

Lungs’ perfusion recovery

(a)

R lobeL lobe

Lungs’ perfusion recovery

(b)

Figure12:3DSPECTLungs’FBPreconstructionbyHannfilter(criticalfrequency0.9)andChangattenuationcorrectionorder0,coefficient

0.11. (a) 4 hours postevent, Right lobe embolism, R lobe volume 0.66lt, and total lungs’ volume 2.85lt. (b) 11 days posttherapy, R lobe

volume 3.28lt, and total lungs’ volume 5.28lt. [37] (modified).

5.FilterSelectionand Standardization

Noise reduction is one of the important tasks in clinical

SPECT imaging. One has to be judicious in the selection of

filters and its parameters for reducing noise, as there may

be some common frequencies in the noise and real image

data. Various digital filters (for reducing noise) have been

proposed.

Butterworth, Gaussian, Hamming, Hanning, and Parzen

are commonly used SPECT filters during FBP reconstruc-

tion, which greatly affect the quality and size accuracy of

image. Salihin Yussoff and Zakaria (2009) [7], in a study by

a cardiac phantom, had selected 272 combinations of filter

parameters and applied on image. Their measurements were

used to calculate contrast, signal-to-noise ratio (SNR), and

defect size. The different filter types produced myocardial

image with different contrast, SNR, and defect size. For

contrast and SNR, Gaussian filter was the best, while Parzen

filterwasthebestinproducingaccuratedefectsize.However,

Butterworth filter was found the best for trade off between

contrast, SNR, and defect size accuracy. Selection of filter

should consider the qualitative or quantitative type of

analysis. For qualitative analysis, high contrast and SNR,

Gaussian filter was suggested. The Butterworth filter was

suggested for quantitative analysis as it is greatly dependent

on both, image quality and size accuracy.

The manufacturers normally supply default filter param-

eters to the user which may not be relevant in different clini-

cal situations. A phantom study was used in filter parameters

standardizationandhasbeencomparedwiththosesuggested

by the vendor. The images were reconstructed using FBP

technique with Chang’s method (attenuation coefficient =

0.125) attenuation correction. A Ramp filter; sixteen differ-

ent Hanning filter, thirteen different Metz filter, and nine

different Butterworth filter parameters were applied during

image reconstruction. Those results did not exactly match

withdefaultones.Thefilterparametersmustbestandardized

before being put to clinical use is the recommendation [41].

FBP reconstruction has been, for a long time, the only

reconstruction algorithm used in SPECT (Figure 13) and

Thyroid coronal slices

RL LL

Figure 13: Sequential coronal slices of Thyroid SPECT study.

Reconstruction made by FBP and Ramp-Butterworth prefiltering.

Right lobe node delineation. Study is presented courtesy of M.

Gavrilelli, MSc, “Medical Imaging Center” Athens, Greece.

is still the reconstruction algorithm recommended for use

in National Electrical Manufacturers Association (NEMA)

performance tests [42].

The Society of Nuclear Medicine in the Guideline for

General Imaging V6.0 9, 2010 gives some recommenda-

tions on prefiltering and reconstruction in SPECT imaging

[43]. Prefiltering of the projection data in SPECT studies

for smoothing in the axial direction must be included.

Reconstruction by FBP demands a ramp filter that corrects

for the smoothing caused by the back projection process.

Filters must be used to restore some of the resolution

lost in the reconstruction process. The particular filter that

is used depends upon the imaging equipment, the depth

of the organ of interest and the radius of rotation. Care

should be taken with image enhancement since it is possible

to produce artifacts. Though, in iterative reconstruction

of SPECT studies the methodology makes it possible to

incorporate correction for many physical effects such as

nonuniform attenuation correction, scatter reduction or

removal, variation of spatial resolution with distance, and so

forth, many times the filtering support is necessary.

Page 11

International Journal of Biomedical Imaging 11

(a)

TOMO FBP coronal

(b)

TOMO IRAC coronal

(c)

TOMO IRAC coronal

(d)

Figure 14: Coronal slices of renal SPECT study. The effect of filtering in smoothing and contrast of SPECT reconstruction, by FBP and

Chang attenuation correction (coefficient 0.11) or OSEM iterations is reflected on clinical images. (a) FBP reconstruction, Butterworth filter

(critical frequency 0.5cm−1, power 10) (b) FBP reconstruction, Hann filter (critical frequency 0.9cm−1) (c) OSEM 10 subsets/10 iterations

(d) OSEM 10 subsets/10 iterations and postfiltering by Butterworth (critical frequency 0.5cm−1, power 10). Study has been completed in

Radiation Physics Unit, Department of Radiology, University of Athens.

6.IterativeReconstruction andFiltering

Iterative image reconstruction methods allow the incorpora-

tion of more accurate imaging models rather than the Radon

model assumed in the FBP algorithm. These include scatter

andattenuationcorrectionsaswellascollimatoranddistance

response and more realistic statistical noise models.

Iterative techniques such as OSEM take into account the

Poisson count distribution and the filters are applied mostly

postprocessing in 3D. Postfiltering with a Butterworth filter

has been shown to result in higher contrasts compared to

reconstructions without filtering (Figures 14(c) and 14(d)).

However, postfiltering with 3D Gaussian filter kernels

should be avoided when collimator detector response com-

pensation is included in the reconstruction. Contrast as a

function of noise has been studied for prefiltering of123IDAT

SPECTimageswith2DGaussianfilterkernelsandtheresults

showed that contrast as a function of noise is comparable for

the prefiltered and nonfiltered OSEM reconstructed images

[44].

OSEM algorithm convergence properties depend on

the activity distribution in the field of view. Resolution

properties for OSEM have been studied [45] with different

types of regularization. Although different parts of the image

converge at different rates, pure and post OSEM filtration

achieve reasonably uniform resolution. Inter-iteration filter-

ing (IF OSEM) with smoothing filters, such as a Gaussian,

produces images with varying spatial resolution that is

dependentonthesurroundingactivity.Itwasconcludedthat

the resolution nonuniformity is entirely due to the filtering.

A spatially varying filter has been proposed to overcome

this problem and to obtain images with nearly uniform

resolution.

Seret in his work [46] in comparison of OSEM and FBP

concludes that one might suggest that the number of subsets

and iterations chosen should be close to the convergence

Page 12

12International Journal of Biomedical Imaging

for all studied regions before quantitative comparisons are

made between FBP and OSEM. The number of requested

iterations will probably result in images that are too noisy,

and a postprocessing filter should be applied.

In a study for comparison of different types of commer-

cialFBPandOSEMSPECTreconstructionsoftware[47]FBP

reconstructions were performed by use of the Ramp filter

limited at the Nq frequency (0.5 cycle per pixel). Prefiltering

of the projections with either the Hanning filter or the

order 3 or 6 Butterworth filter was also considered. Three

cut-off frequencies (0.20, 0.35, and 0.50 cycles per pixel)

were used with the Hanning filter, and 4 cut-off frequencies

(0.10, 0.20, 0.35, and 0.50 cycles per pixel) were applied to

the Butterworth filter. Most of the types of software were

equivalent for FBP or OSEM reconstruction. However, a few

differences were observed with some types of software and

should be considered when they are used.

Comparing four sets of coronal images reconstructed by

FBP or OSEM and different filters (Figure 14) one evaluates

theeffectofprocessingtechniquesandfilteringonthequality

of the image.

Using 3D OSEM with suitable AC may improve lesion

detectability due to the significant improvement of image

contrast. 3D iterative reconstruction algorithms are likely to

replace the FBP technique for many SPECT clinical appli-

cations. Though, more exact image compensation methods

need to be developed and optimal image reconstruction

parametersneedtobeused.Thefullimpactofthesemethods

on quantitative SPECT imaging is yet to be assessed [48].

An efficient postprocessing method to compensate for

scattering and blurring effects in inhomogeneous media was

presented by Yan and Zeng (2008) [49]. The major challenge

of the method is to accurately estimate the 2D point spread

function (PSF) in the image domain. From the clinical

aspect, the implementation of the method is faster than

the iterative reconstruction-based compensation method.

This method is developed in two dimensions and does

not consider scattered photons from out-of-plane sources.

Future work will possibly include modelling the scattering

with a 3D-PSF and a comparison between 3D-PSF method

and 3D-OSEM could be done.

7.DiscussionandConclusion

SPECT has become an important diagnostic tool in nuclear

medicine. SPECT images show characteristic anatomical and

functional information of the structures and the tissues.

The quality of the image depends on several factors such as

spatial resolution (detail or sharpness), contrast and noise

(statistical and structure).

One of the most important factors that greatly affect the

quality of clinical SPECT images is image filtering. Image

filtering is a smoothness process for noise removal and

resolution recovery. A number of filters have been designed

and are available in the reconstruction of tomographic

images. All of them are characterized by two parameters: the

cut-offfrequencyandtheorder.Thesamefilterwithdifferent

parameters can affect variously the image quality. The type

of the filter and the application of the filter parameters

cannot be generalized in all types of clinical SPECT

studies.

The selection of the optimal filter and the determination

offilterparametersforanyindividualcaseremainsoneofthe

mainproblemsoffilteringinSPECTimageprocessing.Espe-

cially the selection of the cut-off frequency is very important

in order to reduce noise and preserve the image details.

Proper filter selection is significant for the improvement of

theimagequalityandthereforeforthediagnostic evaluation.

However, no filter is perfect and there is no specific filter for

all applications. In the literature there are limited reports for

the choice of the appropriate filter parameters in a certain

SPECT examination, as the findings of filter application per

organ reconstruction cannot be generalized.

Nowadays, FBP reconstruction is progressively replaced

with the OSEM-iterative reconstruction algorithm. Unlike

FBP, OSEM is not a linear algorithm, and the reconstructed

contrast depends on the true contrast and on object size.

Moreover, FBP is still faster than OSEM and remains widely

used in clinical practice [40]. Iterative reconstruction meth-

ods have seen a significant growth in tomographic recon-

struction because of the increased computerizing speed.

Iterative reconstruction algorithms produce accurate images

of radioactive distribution and seem to be more sensitive

than FBP technique [47]. Further development in iterative

reconstruction methods will be very promising in improving

image quality. Alzimami et al. (2009) have demonstrated

an improved performance of the new 3-D OSEM method

compared to FBP, particularly for low count statistics. It is

necessary, though, optimal image reconstruction parameters

to be used for the comparison of the full potential of these

methods and evaluation of their impact on quantitative

SPECT imaging [50].

FBP and OSEM are generally both available on all

SPECT processing software developed by gamma camera

manufacturers and the nuclear medicine processing software

companies. The SPECT filters can greatly affect the quality

of clinical images by their degree of smoothing. Proper filter

selection and adequate smoothing helps the physician in

results’ interpretation and accurate diagnosis.

Acknowledgment

The authors wish to thank Ms. Maria Gavrilelli, MSc,

Medical Physicist in Medical Imaging Center, Athens, Greece

for the permission to use her images in this work.

References

[1] P.P.Bruyant,“Analyticanditerativereconstructionalgorithms

in SPECT,” Journal of Nuclear Medicine, vol. 45, no. 10, pp.

1343–1358, 2002.

[2] B. M. W. Tsui, “The AAPM/RSNA physics tutorials for

resident,” Radiographics, vol. 16, no. 1, pp. 173–183, 1996.

[3] K. van Laere, M. Koole, I. Lemahieu, and R. Dierckx, “Image

filtering in single-photon emission computed tomography:

principles and applications,” Computerized Medical Imaging

and Graphics, vol. 25, no. 2, pp. 127–133, 2001.

Page 13

International Journal of Biomedical Imaging 13

[4] S. R. Cherry, J. A. Sorenson, and M. E. Phelps, Physics

in Nuclear Medicine, Saunders, Philadelphia, Pa, USA, 3rd

edition, 2003.

[5] M. W. Groch and W. D. Erwin, “SPECT in the year 2000: basic

principles,”JournalofNuclearMedicineTechnology,vol.28,no.

4, pp. 233–244, 2000.

[6] M. M. Khalil, Ed., Basic Sciences of Nuclear Medicine, Springer,

Berlin, Germany, 2010.

[7] M. N. Salihin Yusoff and A. Zakaria, “Determination of

the optimum filter for qualitative and quantitative

myocardial SPECT imaging,” Iranian Journal of Radiation

Research, vol. 6, no. 4, pp. 173–181, 2009.

[8] E. P. Michael, Ed., PET: Molecular Imaging and Its Biological

Applications, Springer, New York, NY,USA.

[9] M. A. King, S. J. Glick, B. C. Penney, R. B. Schwinger, and

P. W. Doherty, “Interactive visual optimization of SPECT pre

reconstruction filtering,” Journal of Nuclear Medicine, vol. 28,

no. 7, pp. 1192–1198, 1987.

[10] M. A. King, R. B. Schwinger, P. W. Doherty, and B. C. Penney,

“Two-dimensional filtering of SPECT images using the Metz

and Wiener filters,” Journal of Nuclear Medicine, vol. 25, no.

11, pp. 1234–1240, 1984.

[11] J. M. Links, R. W. Jeremy, S. M. Dyer, T. L. Frank, and L. C.

Becker, “Wiener filtering improves quantification of regional

myocardial perfusion with thallium-201 SPECT,” Journal of

Nuclear Medicine, vol. 31, no. 7, pp. 1230–1236, 1990.

[12] G. V. Heller, A. Mann, and R. C. Hendel, Nuclear Cardiology:

Technical Applications, McGraw-Hill, New York, NY, USA,

2009.

[13] A. Takavar, G. Shamsipour, M. Sohrabi, and M. Eftekhari,

“Determination of optimum filter in myocardial SPECT: a

phantom study,” Iranian Journal of Radiation Research, vol. 1,

no. 4, pp. 205–210, 2004.

[14] M. N. Salihin Yussoff and A. Zakaria, “Relationship between

theoptimumcutofffrequencyforButterworthfilterandlung-

heart ratio in99mTc myocardial SPECT,” Iranian Journal of

Radiation Research, vol. 8, no. 1, pp. 17–24, 2010.

[15] P. V´ era, A. Manrique, V. Pontvianne, A. Hitzel, R. Koning,

and A. Cribier, “Thallium-gated SPECT in patients with

major myocardial infarction: effect of filtering and zooming

incomparisonwithequilibriumradionuclideimagingandleft

ventriculography,” Journal of Nuclear Medicine, vol. 40, no. 4,

pp. 513–521, 1999.

[16] P.-Y. Marie, W. Djaballah, P. R. Franken et al., “OSEM

reconstruction, associated with temporal Fourier and depth-

dependant resolution recovery filtering, enhances results

from Sestamibi and 201Tl 16-Interval gated SPECT,” Journal

Nuclear Medicine, vol. 46, no. 11, pp. 1789–1795, 2005.

[17] S. Minoshima, H. Maruno, N. Yui et al., “Optimization

of Butterworth filter for brain SPECT imaging,” Annals of

Nuclear Medicine, vol. 7, no. 2, pp. 71–77, 1993.

[18] E. Raeisi, H. Rajabi, M. R. Aghamiri et al., “Qualitative

evaluation of filter function in brain SPECT,” Iranian Journal

of Nuclear Medicine, vol. 15, no. 27, pp. 1–8, 2007.

[19] M. Lyra, “Single photon emission tomography (SPECT)

and 3D images evaluation in nuclear medicine,” in Image

Processing, Y.-S. Chen, Ed., InTech, Vienna, Austria, 2009.

[20] I. Gunes, I. Sarikaya, T. Ozkan, and T. Akbunar, “Detection

efficiency of different bone SPECT processing protocols for

the diagnosis of “spina bifida”,” Journal of Nuclear Biology and

Medicine, vol. 37, no. 2, pp. 49–52, 1993.

[21] L. E. Holder, J. L. Machin, P. L. Asdourian, J. M. Links,

and C. C. Sexton, “Planar and high resolution SPECT bone

99mTc

imaginginthediagnosisoffacetsyndrome,”JournalofNuclear

Medicine, vol. 36, no. 1, pp. 37–44, 1995.

[22] M. A. King, B. C. Penney, and S. J. Glick, “An-imagedependent

Metz filter for nuclear medicine images,” Journal of Nuclear

Medicine, vol. 29, no. 12, pp. 1980–1989, 1988.

[23] K. Bethge, G. Kraft, P. Kreisler, and P. Walter, Medical

Applications of Nuclear Medicine, Springer, 2004.

[24] J. A. Carrasquillo, J. V. Rogers, D. L. Williams, W. P. Shuman,

D. O. Olson, and S. M. Larson, “Single-photon emission

computed tomography of the normal liver,” American Journal

of Roentgenology, vol. 143, pp. 937–943, 1983.

[25] C. de Sadeleer, A. Bossuyt, E. Goes, and A. Piepsz, “Renal

technetium-99m-DMSA SPECT in normal volunteers,” Jour-

nal of Nuclear Medicine, vol. 37, no. 8, pp. 1346–1349, 1996.

[26] D. Groshar, B. Moskovitz, E. Issaq,and O. Nativ, “Quantitative

SPECT of DMSA uptake by the kidneys: assessment of

reproducibility,” Kidney International, vol. 52, no. 3, pp. 817–

820, 1997.

[27] T.-C. Yen, W.-P. Chen, S.-L. Chang, R.-S. Liu, S.-H. Yeh,

and C.-Y. Lin, “Technetium-99m-DMSA renal SPECT in

diagnosing and monitoring pediatric acute pyelonephritis,”

Journal of Nuclear Medicine, vol. 37, no. 8, pp. 1349–1353,

1996.

[28] M. Lyra, K. Skouroliakou, C. Georgosopoulos, C. Stefanides,

and J. Jordanou, “Single photon emission computed tomogra-

phyand3dimensionalquantitativeevaluationinrenalscintig-

raphy,” in Proceedings of the 4th International Conference on

Medical Image Computing and Computer-Assisted Intervention

(MICCAI ’01), W.J. Niessen and M. A. Viergever, Eds., vol.

2208 of Lecture Notes in Computer Science, pp. 1222–1223,

Springer, London, UK, 2001.

[29] M. Brenner, D. Bonta, H. Eslamy, and H. A. Ziessman,

“Comparison of99mTc-DMSA dual-head SPECT versus high-

resolution parallel-hole planar imaging for the detection of

renal cortical defects,” American Journal of Roentgenology, vol.

193, pp. 333–337, 2009.

[30] N. Sheehy, T. A. Tetrault, D. Zurakowski, A. H. Vija, F.

H. Fahey, and S. T. Treves, “Pediatric99mTc-DMSA SPECT

performed by using iterative reconstruction with isotropic

resolution recovery: improved image quality and reduced

radiopharmaceutical activity,” Radiology, vol. 251, no. 2, pp.

511–516, 2009.

[31] W. R¨ omer, N. Reichel, H. A. Vija et al., “Isotropic reconstruc-

tion of SPECT data using OSEM 3D: correlation with CT,”

Academic Radiology, vol. 13, no. 4, pp. 496–502, 2006.

[32] P. J. Roach, D. L. Bailey, and G. P. Schembri, “Reinventing

ventilation/perfusion lung scanning with SPECT,” Nuclear

Medicine Communications, vol. 29, no. 12, pp. 1023–1025,

2008.

[33] H. Gutte, J. Mortensen, C. V. Jensen et al., “Comparison of

V/Q SPECT and planar V/Q lung scintigraphy in diagnosing

acute pulmonary embolism,” Nuclear Medicine Communica-

tions, vol. 31, no. 1, pp. 82–86, 2010.

[34] P. Reinartz, H. J. Kaiser, J. E. Wildberger, C. Gordji, B.

Nowak, and U. Buell, “SPECT imaging in the diagnosis of

pulmonary embolism: automated detection of match and

mismatch defects by means of image-processing techniques,”

Journal of Nuclear Medicine, vol. 47, no. 6, pp. 968–973, 2006.

[35] B. Harris, D. Bailey, S. Miles et al., “Objective analysis of

tomographicventilation-perfusionscintigraphyinpulmonary

embolism,” American Journal of Respiratory and Critical Care

Medicine, vol. 175, no. 11, pp. 1173–1180, 2007.

Page 14

14International Journal of Biomedical Imaging

[36] P. Weinmann, J.-L. Morettil I, and M. W. Brauner, “Usefulness

of tomographic versus planar lung scintigraphy in suspected

pulmonary embolism in a daily practice,” The Open Medical

Imaging Journal, vol. 2, pp. 49–55, 2008.

[37] M. Lyra, M. Gavrilelli, V. Lyra, G. Kokona, and K. Skouro-

liakou, “Lungs SPECT image processing for volume and

perfusion index estimation,” in Proceedings of the 8th IEEE

International Conference on BioInformatics and BioEngineering

(BIBE ’08), pp. 1–5, Athens, Greece, October 2008.

[38] H. Zaidi, “Comparative methods for quantifying thyroid

volume using planar imaging and SPECT,” Journal of Nuclear

Medicine, vol. 37, no. 8, pp. 1421–1426, 1996.

[39] J. W. van Isselt, J. M. H. de Klerk, P. P. van Rijk et al.,

“Comparison of methods for thyroid volume estimation in

patients with Graves’ disease,” European Journal of Nuclear

Medicine and Molecular Imaging, vol. 30, no. 4, pp. 525–531,

2003.

[40] Y.-W. Bahk, S.-K. Chung, Y.-H. Park, S.-H. Kim, and H.-K.

Lee, “Pinhole SPECT imaging in normal and morbid ankles,”

Journal of Nuclear Medicine, vol. 39, no. 1, pp. 130–139, 1998.

[41] A. K. Pandey, G. S. Pant, and A. Malhotra, “Standardization of

SPECT filter parameters,” Indian Journal of Nuclear Medicine,

vol. 19, no. 2, pp. 30–35, 2004.

[42] Performance Measurement of Scintillation Cameras, National

Electrical Manufacturers Association. NEMA Standards Pub-

lication, Rosslyn, Va, USA, 2001.

[43] J. A. Parker et al., “Task Force Members (V6.0),” SNM

Procedure Guideline for General Imaging, 2010, http://inter-

active.snm.org/docs/General Imaging Version 6.0.pdf.

[44] A. Larsson, S. J. Mo, T. Sundstr¨ om, and K. Riklund, “Gaussian

pre filtering of 123I DAT SPECT images when using depth-

independent resolution recovery,” Physics in Medicine and

Biology, vol. 52, no. 18, pp. N393–N399, 2007.

[45] S. Mustafovic, K. Thielemans, D. Hogg, and P. Bloomfield,

“Object dependency of resolution and convergence rate in

OSEM with filtering,” in Proceedings of the Nuclear Science

Symposium Conference Record, vol. 3, pp. 1786–1790, Novem-

ber 2001.

[46] A. Seret, “Number of iterations when comparing MLEM/

OSEM with FBP,” Journal of Nuclear Medicine, vol. 45, no. 12,

p. 2125, 2004.

[47] A. Seret and J. Forthomme, “Comparison of different types

of commercial filtered backprojection and ordered-subset

expectation maximization SPECT reconstruction software,”

Journal of Nuclear Medicine Technology, vol. 37, no. 3, pp. 179–

187, 2009.

[48] S. Vandenberghe, Y. D’Asseler, R. van de Walle et al., “Iterative

reconstructionalgorithmsinnuclearmedicine,”Computerized

Medical Imaging and Graphics, vol. 25, no. 2, pp. 105–111,

2001.

[49] Y. Yan and G. L. Zeng, “Scatter and blurring compensation

in inhomogeneous media using a post processing method,”

International Journal of Biomedical Imaging, vol. 2008, Article

ID 806705, 11 pages, 2008.

[50] K. S. Alzimami, S. A. Sassi, and N. M. Spyrou, “A comparison

between 3D OSEM and FBP image reconstruction algorithms

in SPECT,” in Advances in Electrical Engineering and Computa-

tional Science, vol. 39 of Lecture Notes in Electrical Engineering,

pp. 195–206, Springer, 2009.

#### View other sources

#### Hide other sources

- Available from Maria Lyra · May 28, 2014
- Available from PubMed Central
- Available from hindawi.com
- Available from hindawi.com