Low dose CT for attenuation correction in PET. Validation of quantification for different patient sizes.

in PET. Validation of quantification for

different patient sizes.

Anders Törnblom

Master Thesis in Medical Radiation Physics

Stockholm University

supervised by Martin Bolin

MSc: Dept. Nuclear Medicine, Karolinska University Hospital Jonathan Siikanen

Ph.D: Dept. Nuclear Medicine, Karolinska University Hospital Daniel Thor

Abstract

Introduction: Despite the relatively low dose (0.5 mSv – 1 mSv) generated by Attenuation Correction CT (ACCT) in PET examinations, the ALARA principle is still applicable. The currently used ACCT standard protocol at the Karolinska University Hospital in Solna uses 7.6 effective mAs (mAseff) and 120 kVp, but reduc-ing mAseff and/or kVp would decrease patient dose as well as facilitate an increased number of research subjects. A CT reconstruction algorithm called Quantification Achieved Consistently (Q.AC.) (Lonn, 2012) has recently been developed to enable reduced doses from ACCT, while preserving quantitative PET data. The purposes of this study were to investigate possible limitations of the Q.AC. with respect to patient size, and to optimise protocols, aiming at minimising ACCT dose in terms of Volumetric Computer Tomography Dose Index CT DIvol.

Methods: Measurements were performed with a GE PET/CT Discovery sys-tem, which offers Q.AC. reconstruction. The NEMA NU-2 protocol was followed to quantify PET quality, including evaluations of relative count error in the arti-ficial lung in the phantom centre (∆lung), hot- and cold-sphere contrast (Q), and background variability (N ). Two phantoms were used; the NEMA body phantom (elliptical cross section sized 30 cm laterally and 23 cm anterior-posterior (AP)), here representing paediatric patients and small-sized adults, and the same phantom with an additional (20 cm laterally and 4 cm AP) ellipsoid plastic (PMMA) exten-sion ring, representing mid- and large-sized patients. ACCTs were acquired with 15 mAseff values, range [2.3 - 260], in combination with four kVp values [80, 100, 120, 140] and reconstructed with two algorithms (Q.AC. and a regular soft CT algo-rithm). Consequently, PET reconstructions were performed based on each mAseff, kVp and CT-reconstruction combination.

Results: Quantitatively similar PET results to the standard protocol were achieved with the Q.AC. CT reconstruction algorithm using a CT DIvol = 0.06 mGy (2.3 mAseff and 80 kVp) for the NEMA body phantom, respectively a CT DIvol = 0.20 mGy (2.3 mAseff and 120 kVp) for the phantom with additional extension ring.

Conclusions: This study indicates that the Q.AC. CT reconstruction algorithm enables accurate PET results at lower ACCT mAseff and kVp settings than the currently used clinical standard protocol. For paediatric patients and small-sized adults, a reduction of CT DIvol by approximately 90% may be achieved, while for mid- and large-sized patients, the CT DIvol can be reduced by approximately 70% without loss of quantitative PET data.

Contents

1 Introduction 7

1.1 Nuclear Medicine and Tomography . . . 7

1.2 Attenuation and Scatter . . . 8

1.3 This project . . . 8

2 Theory 9 2.1 Photon Interactions . . . 9

2.1.1 Interaction processes for photon energies ≤ 511 keV in tissue . . . 10

2.1.2 Compton Scattering in PET . . . 10

2.2 Radiation Doses and ALARA . . . 11

2.3 Computed Tomography (CT) . . . 12

2.3.1 Basic Principles . . . 12

2.3.2 CT Reconstruction . . . 13

2.3.3 CT Acquisition Settings . . . 14

2.3.4 CT dosimetry . . . 16

2.4 Positron Emission Tomography (PET) . . . 17

2.4.1 Basic Principles . . . 17

2.4.2 Detectors . . . 18

2.4.3 Spatial Resolution . . . 18

2.4.4 Time-Of-Flight TOF . . . 19

2.4.5 Attenuation and Scatter Correction in PET . . . 20

2.5 Using CT Data in PET Analyses . . . 22

2.5.1 Attenuation Correction CT (ACCT) . . . 22

2.5.2 Iterative Reconstruction . . . 24

2.5.3 Quantitation Achieved Consistently – Q.AC. . . 25

3 Material and Methods 27 3.1 Investigation Procedure . . . 27

3.2 Phantoms . . . 28

3.3 NEMA NU 2-2012 – Performance Measurement of PET Systems . . . 29

3.4 Instrument Settings (Measurement Data collection) . . . 31

3.4.1 ACCT Acquisition Settings . . . 31

3.4.2 PET Acquisition Settings . . . 32

3.5 Data Analysis . . . 32

4 Results 35 4.1 Measured Activity in Lung Insert ∆lung . . . 35

4.2 Measured Contrast in Cold Spheres QC . . . 38

4.3 Measured Contrast in Hot Spheres Q_H . . . 41

5 Discussion 47

5.1 Measured Activity in Lung Insert ∆lung . . . 47

5.2 Measured Contrast in Cold Spheres QC . . . 47

5.3 Measured Contrast in Hot Spheres Q_H . . . 48

5.4 Measured Background Variability N . . . . 48

5.5 Summary . . . 49 6 Conclusion 51 7 Outlook 51 A Figures 56 B Error calculations 59 C Protocol 60 C.1 PET Data Aqcuisition Protocol . . . 60

List of Figures

2.1 Illustration of three photon interactions: Photoelectric Effect (PE) 1: Pri-mary photon and 2: Characteristic X-ray, Compton Scattering (CS), and Pair Production (PP) . . . 9

2.2 Photon interaction type depending on photon energy and the number of protons per atom (Z) of the attenuating material. The curved lines indicate where the main mode of interaction changes. In the body the highest Z-material, with significant prevalence, is ₂₀Ca and the photon energy is approximately 0.5 MeV in PET. The red rectangle cover the approximate region within which the relevant combinations of photon energies and Z seen in PET imaging are found. I.e. Compton scatter is the dominant mode of interaction. Adapted from (Podgorsak, 2010) . . . 10

2.3 Helical CT acquisition. . . 13

2.4 An example of the performance steps in Filtered Back Projection (FBP) . 14 2.5 X-ray spectra generated with, 80, 100, 120, 140 kV respectively on a tungsten target. The dashed lines are the respective unfiltered spectra. (Cattin, 2016). Used with permission from Prof. Dr. Philippe C. Cattin. . 15

2.6 a) True coincidence: Annihilation photons reach respective detector with-out interacting generating true LOR. b) Scatter coincidence: One of the two photons is scattered generating erroneous LOR (red line). c) Ran-dom coincidence: Two annihilation events coincides where one of each annihilation event photon misses the detector generating erroneous LOR (red line). . . 17

2.7 Illustration of the variation of the apparent detector width due to off-centre location of the LOR. . . 19 2.8 Illustration of annihilation photons, originating off-centre, traversing air.

See eq. 2.11 . . . 20 2.9 Two annihilation photons traversing a homogeneous material with

atten-uation = µ surrounded by air. Schematic illustration of eq. 2.13 . . . 21 2.10 Qualitative illustration of the impact of AC and SC for quantitative PET

imaging of an object with a homogeneous Activity Concentration (C_A) (without noise). . . 22 2.11 Bilinear HU-to-µ_511keVconversion plot. (John R. Votaw, 2015) . . . 23 2.12 Example of ACCT to attenuation map conversion: 140 kVp and 775 mAs_eff

ACCT (a) converted to an attenuation map (b). . . . 23 2.13 Flowchart of the iterative reconstruction algorithm MLEM. . . 24 2.14 CT acquisition with 80 kVp and 2.3 mAs_eff. CT reconstructions using, a)

regular soft CT algorithm, and b) the Q.AC. algorithm. Pixels units [HU]. Attenuation maps µ_511keV based on CT reconstruction using, c) regular soft CT algorithm, and d) the Q.AC. algorithm. Pixel unit [cm2/g] . . . . 26 3.1 Photo of the PET/CT Discovery MI DR used for acquisition . . . 27 3.2 NEMA Body Phantom . . . 28 3.3 Photo of the extension ring made of PMMA (dimensions: 50 cm laterally

and 27 cm AP). It surrounds the NEMA body phantom to increase the attenuation and this simulated a medium/large sized patient. . . 29 3.4 Anthropomorphic phantom filled with water and cylindrical spine insert.

Two legs were positioned beside the phantom to simulate clinically rele-vant attenuation. . . 31 3.5 ROIs used in the data analysis of the PET images. . . 33 3.6 CT template for reslicing and rescaling PET images prior to data analysis. 34 4.1 NEMA Body Phantom Results ∆_lung vs. mAs_eff. . . 36 4.2 NEMA Body Phantom with Extension Ring Results ∆_lung vs. mAs_eff. . . 37 4.3 NEMA Body Phantom QC vs. mAseff Results: 37 mm sphere comparing

Q.AC. to Regular soft CT algorithm. . . 39 4.4 NEMA Body Phantom with extension ring Q_C vs. mAs_eff Results: 37

mm sphere comparing Q.AC. to Regular soft CT algorithm. . . 40 4.5 NEMA Body Phantom QH vs. mAseffResults: 10 mm and 22 mm sphere

comparing Q.AC. to Regular soft CT algorithm. . . 42 4.6 NEMA Body Phantom with extension ring QH vs. mAseff Results: 10

mm and 22 mm sphere comparing Q.AC. to Regular soft CT algorithm. . 43 4.7 NEMA Body Phantom N vs. mAs_eff Results: 10 mm and 37 mm sphere

comparing Q.AC. to Regular soft CT algorithm. . . 45 4.8 NEMA Body Phantom with extension ring N vs. mAseff Results: 10 mm

and 37 mm sphere comparing Q.AC. to Regular soft CT algorithm. . . 46 5.1 Contrast Q_H and Q_C result for the lowest CT DI_vol value of respective

5.2 Background Variability N results for the lowest CT DI_vol value of respec-tive phantom size and QA measurements results for comparison. . . 51 A.1 NEMA body phantom with Measurement results with with varying mAseff

using the Q.AC. CT reconstruction algorithm. . . 56 A.2 NEMA body phantom with extension ring: Contrast Q measurement

re-sults with with varying mAseffusing the Q.AC. CT reconstruction algorithm. 57

A.3 NEMA body phantom with extension ring: Background Variability N measurement results with with varying mAs_eff using the Q.AC. CT re-construction algorithm. . . 58 C.1 Data acquistion protocol (page 1) . . . 60 C.2 Data acquistion protocol (page 2) . . . 61

List of Tables

3.1 List of the different mAseff to investigate in combination concluded from

the anthropomorphic water phantom investigation. . . 31 5.1 Lowest possible CT DI_vol achieving optimal quantitative-PET quality for

List of abbreviations

• ACCT: Attenuation Correction Com-puted Tomography

• ALARA: As Low As Reasonably Achievable • AP: Anterior-Posterior • C_A: Activity Concentration • CS: Compton Scattering • CT or TCT: Transmission Computed Tomography

• CTDI: Computed Tomography Dose Index

• ◦: Diameter

• mAseff: Effective mAs

• ET or ECT: Emission Computed To-mography

• FBP: Filtered Back Projection • FDG: Fluorodeoxyglucose

• FWHM: Full Width at Half Maxi-mum

• HU: Hounsfield Unit

• ICRP: International Commission on Radiological Protection

• LNT: Linear NonThreshold • LOR: Line of Response

• LSO: Lutetium Oxyorthosilicate (Lu2SiO5(Ce))

• LySO: Lutetium Yttrium Oxy-orthosilicate

• MLEM: Maximum Likelihood Esti-mation Maximization

• NEMA: National Electrical Manufac-turers Association

• OSEM: Ordered Subset Estimation Maximization

• PET: Positron Emission Tomography • PE: Photoelectric Effect

• PMMA: Poly(methyl methacrylate) • PP: Pair Production

• PVE: Partial Volume Effect • QA: Quality Assurance

• Q.AC.: Quantitation Achieved Con-sistently

• ROI: Region Of Interest • SC: Scatter Correction • SNR: Signal-to-Noise Ratio

• SPECT: Single Photon Emission To-mography

• SSM: Svenska StrålsäkerhetsMyn-digheten (Swedish Radiation Safety Authorities)

1

Introduction

1.1 Nuclear Medicine and Tomography

George de Hevesy’s seminal publication on radio-traces in plants from the spring of 1923 was pivotal for what became the field, currently, known as diagnostic nuclear medicine (Hevesy, 1923). In diagnostic nuclear medicine, biological processes (e.g. cell metabolism and/or organ function) are traced through either replacing stable isotopes with radioac-tive isotopes (radioisotopes), or by inserting radioisotopes into molecules both that are involved in the biological processes to be studied. Radioisotopes or molecules carry-ing a radioisotope are referred to as tracers, which are used to non-invasively study aforementioned processes. With a detector (camera), it is possible to locate the tracer concentration within an object, by measuring the emitted photons yielded by the ra-dioisotopes.

Tomography is an imaging technique to non-invasively image the inside of an ob-ject. This is possible with the mathematical theory called Radon transform where a 3D volume can be reconstructed based on 2D projections (shadows) through an object. Two tomographic imaging techniques used in medicine are subjects of this thesis. (1) Injection of a tracer into an object measuring photons emitted from within the object with an external detector, so-called Emission Computed Tomography abbreviated to ET. (2) External radiation transmission of an object, measuring the residual radiation with a detector on the opposite side of the object, so-called Transmission Computed Tomography abbreviated to CT. Illustrations of ET and CT are shown in figure 1.1.

CT

Emission

Tomography

Photon Detector Photon

source Object

⇒

⇐

ma

Figure 1.1: The difference between Emission Tomography (ET) and Computed Tomography (CT).

The Positron Emission Tomography (PET) camera is a diagnostic tool that utilizes the unique property of proton rich radioisotopes that decay and generate positrons (β+

When a positron annihilates with an electron, two anti-parallel photons are emitted. By recording large numbers of these annihilation photons, the activity distribution within an object can be determined i.e. the physiological function of the tracer.

The first (PET) systems (PC-I and PC-II) for human investigation were built in 1974 (Phelps et al., 1975) and a few years later, based on these systems, the first commercial PET systems were developed by the Cyclotron Corporation Inc. (Brownell, 1999).

CT measures the electron density throughout an object by irradiating it with a spectrum of photons which originate from an X-ray tube. The intensity of the photons is measured on the opposite side with a detector. The X-ray tube and detector are rotated around the object acquiring several projections.

The first commercial CT scanner came 1973 (Hounsfield, 1973; Ambrose, 1973), which only fitted a head but was later further developed to fit a body.

1.2 Attenuation and Scatter

A problem encountered when backtracking tracer origins through measurements of the annihilation photons, is photon interactions with tissue. These yields either attenuation (loss of signal) or scatter (distorted signal) and degrades the image quality. In order to obtain a quantitative image with PET, these effects have to be compensated for.

Throughout the history of ET, various methods to determine attenuation maps (a transmission scan with data of the magnitude of attenuation in each point of an object) have been developed. E.g. in PET, a rod with 68_Ge/68_{Ga, emitting positrons, was}

externally rotated around the patient to determine the attenuation through an object from several angles, like a mono-energetic CT (Zaidi and Hasegawa, 2003). This method generated high quality attenuation maps but was both complicated to incorporate into the examination process and time consuming due to the necessity of acquiring the at-tenuation map before radio-tracer injection. In 2001, a great technological development for PET imaging came when the first commercial combined PET/CT were developed (Beyer et al., 2000). One of the benefits was the allowance for fast and simple Scatter Correction (SC) and Attenuation Correction (AC) based on CT data, called Attenuation Correction CT (ACCT).

However, with the transition to ACCT, the radiation doses increased compared to e.g. the external rod method (Wu et al., 2004), hence, increased the risk of developing cancer or hereditary diseases, which is further discussed in section 2.2. Therefore, the CT settings and the reconstruction algorithms for generating ACCT should be optimized to minimize the accompanying dose while preserving the quantitative quality of the PET image.

1.3 This project

A white paper from GE Healthcare published in 2012 describes a new ACCT recon-struction algorithm called Quantification Achieved Consistently (Q.AC.) that promises consistent ACCT quality with CT parameter settings corresponding to a significantly lower ACCT dose than previously possible (Lonn, 2012).

The subject of this thesis was to investigate the possibilities to further optimize the acquisition of ACCT, by incorporating, and push the limits of, the Q.AC. CT recon-struction algorithm. Evaluations have been made through quantification of PET images following the National Electrical Manufacturers Association (NEMA) NU 2-2012 proto-col (NEMA, 2013) for two different sized phantoms representing paediatric/small-sized adult patients, and medium/large patients respectively.

2

Theory

2.1 Photon Interactions

The atom has a positively charged nucleus, of protons and neutrons, that is surrounded by shells with negatively charged electrons. A photon is an electromagnetic wave that is considered to have both the properties of a wave and a particle and can interact with atoms.

There are three main types of photon interaction: (1) Photoelectric Effect (PE) where the photon is absorbed to eject an electron; if the electron is replaced by an electron from an outer shell, a secondary photon is emitted, a so-called characteristic X-ray. (2) Compton Scattering (CS) where the photon transfers a fraction of its energy to an electron releasing it from the atom and changes trajectory (scatter). (3) Pair Production (PP) where the photon interacts with the nucleus and is converted into an electron and a positron (anti-electron). The described photon interaction are illustrated in figure 2.1.

The outcome (mode of interaction) of said interactions varies with photon energy (Eγ) and depends on the number of protons (Z) the atom is composed of. The

to-tal probability of all interaction is denoted with µ [m−1] called the linear attenuation coefficient.

Pair Production Photoelectric Effect Compton Scattering

Positron Electron Proton Neutron Photon 2 1

§

Mmm

a.

iodide

~

w•e¥

,

Figure 2.1: Illustration of three photon interactions: Photoelectric Effect (PE) 1: Primary photon and

2.1.1 Interaction processes for photon energies ≤ 511 keV in tissue

Soft tissue is mainly composed of the elements;₁H,₆C,₇N,₈O, which means that the Z in tissue varies between Z ∈ [1, 8]. Skeletal bone is mainly composed of ₂₀Ca,₁₅P, and

8O but is relatively less prominent compared to soft tissue.

Photons in PET imaging are annihilation photons with an energy of 511 keV, which are further described in section 2.4.1. The probability of a specific mode of interaction as a function of Z vs. Eγ is illustrated in figure 2.2 adapted from (Podgorsak, 2010),

there, the approximate region relevant to PET imaging is marked with a rectangle. The combination of a low effective Z in tissue and the relative high energy of annihilation photons makes the probability of PE as a primary interaction low. 511 keV is also too low to achieve PP where the photon must have a minimum energy corresponding to the rest mass of an electrons and a positron (1022 keV) i.e. for the majority of isotopes used in PET, the probability of PP is low. Therefore, the most prominent mode of interaction is CS.

Pair Production

Compton

Scattering

Photoelectric

Effect

0.5

1

5

10

50 100

Photon Energy [MeV]

20

0

40

Z of

Absorber

80

100

120

÷

-=

✓

L

-=

Figure 2.2: Photon interaction type depending on photon energy and the number of protons per

atom (Z) of the attenuating material. The curved lines indicate where the main mode of interaction changes. In the body the highest Z-material, with significant prevalence, is20Ca and the photon energy

is approximately 0.5 MeV in PET. The red rectangle cover the approximate region within which the relevant combinations of photon energies and Z seen in PET imaging are found. I.e. Compton scatter is the dominant mode of interaction. Adapted from (Podgorsak, 2010)

2.1.2 Compton Scattering in PET

In CS, a part of the photon energy is transferred to an electron resulting in; change photon trajectory of angle θ, loss in photon energy, and ejection of the electron from the

atom. The effects of CS in PET image quality are either; attenuation: photon misses the detector i.e. decreased beam intensity, or scatter: photon hits the detector ring but with less energy than when it originated and/or hit incorrect detector element. The resulting energy of a Compton scattered photon as a function of a scatter angle can be expressed as: ECS= E0 1 + E0 511 keV(1 − cos θ) . (2.1)

Here, ECS is the photon energy of the scattered photon, E0 is the initial photon energy,

511 keV is the rest mass (expressed in energy) of an electron and θ is the angle with which the photon trajectory diverges from the initial trajectory (Podgorsak, 2010).

Photons originating in the vicinity of an objects centre, the probability of attenuation and scatter (µ) is higher, yielding a decreased intensity compared to photons emitted in the vicinity of an objects surface with an equal initial intensity. This is one of the major problems in emission tomography and needs to be corrected for in order to ob-tain quantitative images. If the attenuation throughout the object is determined, the photon intensity in each point can be amplified with respect to the attenuation in the corresponding point.

2.2 Radiation Doses and ALARA

Ionizing radiation are electromagnetic waves (photons) or particles which carry enough energy to ionize matter, including tissue. Ionizing radiation in tissue can generate di-rect DNA damages, or reactive oxygenation species, also known as free radicals (highly reactive molecules), that can break DNA strands and/or fixate the DNA strand dam-ages. The outcome of ionizing radiation is magnitude dependent and can generate both stochastic risks and deterministic effects. The quantity of induced ionizing radiation to matter is called dose and is measured in units of Gray [Gy] defined as the energy imparted per unit mass by ionizing radiation to matter at a specific point [J/kg]. To efficiently relate the stochastic effects of a certain dose, a radiation protection quan-tity was developed. This quanquan-tity is mainly based on atomic bomb survivor data and is called effective dose. The effective dose is measured in units of Sievert [Sv], named after the Swedish physicist Rolf Sievert (ICRP 26 1977). The effective dose accounts for the type of radiation, and what part or parts of the body/organ it was deposited in. The increased stochastic risk of developing secondary long-term effects (lethal cancer or hereditary diseases) is approximately 5.7% per Sv for doses above 100 mSv (ICRP 103 2007). However, radiation protection regulations authorities call for the use of the so-called Linear Non-Threshold (LNT) model to estimate the increased risk of secondary long-term effects of radiation in medicine. This model assumes all dose (even below 100 mSv) increase the risk of secondary long-term effects (Cherry, Sorenson, and Phelps, 2012).

Ionizing radiation in medicine, such as in diagnostics (X-ray) and nuclear medicine examinations, increases the risk of stochastic effects but these should be outweighed by the medical benefit of the examination. There are no dose limits to patients or test

sub-jects. However, to minimize the dose burden on patients, radiation-protection boards e.g. International Commission on Radiological Protection (ICRP) and Swedish Radia-tion Safety Authorities (Svenska Strålsäkerhetsmyndigheten (SSM)) aspire for clinics to incorporate the ALARA-principle (As Low As Reasonably Achievable) into work pro-cedures involving ionizing radiation. ALARA state that all justified propro-cedures should be optimized to deposit lowest dose possible without compromising the quality of the procedure.

For clinical purposes, the dose contribution to the total dose of an Attenuation Correction CT (ACCT) is low (0.5 mSv – 1 mSv) compared with diagnostic CT (2 mSv – 7 mSv) and diagnostic PET radio-tracer (2 mSv – 6 mSv) (SSM, 2018). However, with advancing technology the sensitivity of PET-systems increases, and radio-tracer doses can be reduced while preserving the PET-image quality. E.g. the total-body PET, in development, claims to increase the sensitivity 4-5 times (Cherry et al., 2018) allowing for significant reduction of PET radio-tracer doses. Such increments increase the significance of the ACCT dose contribution and will be important to reduce. 2.3 Computed Tomography (CT)

2.3.1 Basic Principles

To generate X-rays, electrons are accelerated over a potential difference U [V] in vacuum hitting a solid metal-target where they are decelerated by coulomb forces, which yield bremsstrahlung (X-rays) i.e. a range of photon energies. Electrons that transverse the potential difference will gain a kinetic energy equal to the potential difference i.e. U = x kV → E_e−= x keV.

A planar X-ray imaging system measures the attenuation, through an object gener-ating a forward projection (planar image). A CT acquires several forward projections by rotating an X-ray tube and a detectors around the object during the exposure, generat-ing a sinogram1 with all forward projections that can be backwards projected to give a tomographic image of the object.

Depending on the patient size and age, the CT acquisition is modified to minimize the radiation dose relative the diagnostic purpose e.g. X-ray tube current (I) and voltage (U ). There are several acquisitions techniques and parameters which can be modified to optimize an acquisition. The most common technique to image larger anatomical regions is the spiral acquisition, where the couch moves transversely (along z-axis) simultaneous with the X-ray tube rotating, also known as helical acquisition, see figure 2.3.

1

Sinogram is a data-structure form resulting from performing Radon transformation of a 3D object. Each projection is stored in a 2D array, with pixel data (intensity and position information), that is associated with an angle related to rotation around a fix point. In the sinogram, the arrays are ordered according to pixel position relative the centre of rotation, and acquisition angle (often denoted

EEEE.EE#Q

Figure 2.3: Helical CT acquisition.

2.3.2 CT Reconstruction

Direct back-projection (reconstruction) of the sinogram (acquired projections) results in an image that appears to be smeared and lacks details e.g. well-defined anatomical structures.

In the early 1970’s, methods to counteract these effects were investigated and at the forefront were the two scientist duos Ramachandran and Lakshminarayanan (Ra-machandran and Lakshminarayanan, 1971), and Shepp and Logan (Shepp and Logan, 1974), which led to the development of the, to date well-established, method referred to as Filtered Back Projection (FBP), where a so-called high-(frequency)-pass filter is convolved with the sinogram in the frequency domain. The filtered sinogram is then inversely Fourier transformed back to the spatial domain. This dampens low frequencies (large structures) and amplifies high frequencies (small structures), generating a clear image when back projected.

However, the FBP method is not strictly for high-pass filters. The filter can be ap-plied either in the spatial or frequency domain and shaped to supress/amplify either high or low frequencies, to fit the acquisition purpose e.g. imaging bone (high frequencies) or soft tissue (low frequencies). In figure 2.4, an illustration of the steps in performing a FBP is shown.

Sinogram of acquired projection data FT High-pass filter Sinogram f f [Hz] A _A Filtered Sinogram A f [Hz] FT-1 r r Filtered sinogram Back project

FT: Fourier transform : Convolution : Angle [cm] [cm] [°] [°] A: Amplitude

"

imam

:

÷

:*

.

.

Figure 2.4: An example of the performance steps in Filtered Back Projection (FBP)

2.3.3 CT Acquisition Settings

In this section, acquisition parameters relevant to CT quality and dose are presented. • Kilo Voltage Peak – kVp

The kVp refers to the maximum photon energy in the X-ray-spectrum, which also will be equal to the electron energy, E_e− = x keV → E_γ,max = x keV. See example of X-ray spectra produced with a tungsten target for various kVp in figure 2.5 (Cattin, 2016).

Figure 2.5: X-ray spectra generated with, 80, 100, 120, 140 kV respectively on a tungsten target. The

dashed lines are the respective unfiltered spectra. (Cattin, 2016). Used with permission from Prof. Dr. Philippe C. Cattin.

• Pitch Factor – p

For helical CT the pitch is defined as the ratio of table translation per 360◦ rotation of the X-ray tube relative to the nominal beam width (N_sT ) of a helical CT (Dance et al., 2014).

p = l

NsT

. (2.2)

Where l is the length of table movement per 360◦ X-ray tube rotation, Ns is

the number of slices acquired during X-ray exposure and T is the nominal slice thickness.

• Rotation Speed – t_rot

The rotation speed, denoted trot is defined as the time it takes for the X-ray tube

to revolve 360◦.

• mAs and Effective mAs (mAseff )

mAs refers to the product of beam current I [A] = [C/s] and the time that the current is applied the charge shot at the target. For a CT acquisition, t_rot[s], pitch factor, and current are set. Then the time the current applied is given as the ratio of t_rot and pitch and the mAs_eff can be determined with eq. (2.3).

mAseff =

trot

p · I = t · I. (2.3)

The investigated PET/CT modality applies fixed values of mA (minimum 10 mA and increases in steps of 5 mA), and the mAs_eff may also be altered by varying t_rot and p. According to eq. (2.3), to reduce mAs_eff, t_rot is decreased and p is increased. In this study the mAseffranges between 2.3 mAs to 260 mAs The mAseff is directly proportional

2.3.4 CT dosimetry

As discussed in section 2.2, the dose is measured in Gray [Gy], which is energy deposited per unit mass [J/kg]. The dose to the patient in a CT examination depends on patient size, mAseff, and kVp. To estimate the dose, an international standardized quantity has

been established called Computer Tomography Dose Index CT DI, which is essential for the calibration of the CT system’s built-in dose monitor. See eq. (2.4) (Hsieh, 2015).

CT DI = 1

NsT

Z z0

−z0Da

(z)dz. (2.4)

Here, Da(z) is the absorbed dose distribution along the z-axis in a stationary CT

ac-quisition, and NsT is nominal beam width (previously defined in section 2.3.3). The

dosimetry measurements for calibration of the CT system’s built-in detectors are per-formed with a cylindrical Poly(methyl methacrylate) (PMMA) plastic phantom with one cylindrical dosimeter placed along its central axis and several cylindrical dosimeters placed inside of the phantom periphery (10 mm from surface). The dose distribution is determined from z = −50 mm to z = +50 mm over the phantom, resulting in the so-called CT DI100 see eq. (2.5).

CDT I100= 1 NsT Z 50mm −50mm Da(z)dz. (2.5)

The results from the CT DI₁₀₀ measurements in the phantom centre and periphery are combined into a weighted CT DI denoted CT DIw see eq. (2.6). The dose in the

periph-ery typically ranges from 3 to 5 times higher compared to the centre (Hsieh, 2015) and thus, the peripheral CT DI₁₀₀ is given a larger weight.

CT DIw =

1

3(CT DI100,c+ 2CT DI100,p) . (2.6) Here, CT DI_100,cis CT DI₁₀₀at the centre of the scanned reference phantom and CT DI_100,p is the mean CT DI100 of the peripheral measurements.

CT DIw is used as a reference to estimate the dose for CT acquisition using other

parameter settings than used for calibration and is the called CT DI_vol defined as:

CT DIvol= nCT DIw,refItrot p U Uref !n kOB. (2.7)

Where_nCT DIw,ref is the normalized and weighted CT DI per mAs for a specific kVp and

total beam collimation, U_ref is the tube voltage used to determine the_nCT DIw,ref, n is

approximately 2.5 (depending on X-ray spectrum) and kOB is the over beaming factor,

i.e. how much was irradiated without contributing to the image due to the detector area being smaller than the beam width.

Finally, to estimate the total amount of radiation used in CT examination, the quantity called Dose-Length Product DLP [Gy·cm] is determined (Huda and Mettler, 2011) defined as the product of the total body length irradiated L and CT DI_vol.

However, this study focuses on the CT DI_vol since it is the common denominator for estimation of the dose in intra-subject acquisitions and independent of the scan length. 2.4 Positron Emission Tomography (PET)

2.4.1 Basic Principles

As described in section 1.1, PET is an imaging modality that maps the distribution of radio-tracers emitting positrons. When a positron collides with an electron, the two particles annihilate resulting in the emission of two anti-parallel (±0.5◦) photons each with an energy of 511 keV (corresponding to the rest mass of an electron or a positron). Simultaneous detection, within a coincidence time window (∼ 4 ns), of these photons enables the measurement of the radio-tracer distribution inside e.g. a patient.

The PET-camera design is a ring with detectors mounted on its circumference di-rected toward its centre. Photon registration in two detectors within the coincidence window creates a so-called Line-Of-Response (LOR) (see figure 2.6 (Cherry, Sorenson, and Phelps, 2012)) between the two detectors, indicating that an annihilation event oc-curred somewhere on that line. The annihilation photons are emitted isotropically and therefore, with multiple annihilations, the sum of LORs generates a focal point that corresponds to the radio-tracer source.

However, a large fraction of the annihilation photons interacts with the tissue through e.g. Compton Scattering (CS) i.e. become attenuated and/or scattered, which may require corrections that are further discussed in section 2.4.5.

a)

b)

c)

.

i÷÷.n¥*

.iq#nniwi

.

onion .in#..&.im'the.iq.

ifeng.n.pro.io:9674 .

÷gw¥*¥i÷a.int.

{

"

gd

Figure 2.6: a) True coincidence: Annihilation photons reach respective detector without interacting

generating true LOR.

b) Scatter coincidence: One of the two photons is scattered generating erroneous LOR (red line). c) Random coincidence: Two annihilation events coincides where one of each annihilation event photon

misses the detector generating erroneous LOR (red line).

Contrary to CT, which images anatomy, PET images physiological function by means of the biological functionality of the radio-tracer used in the examination.

One example of a radio-tracer is FDG (Fluorodeoxyglucose) where18F is chemically attached to a modified glucose molecule. The FDG is diluted in saline and injected intravenously, transported throughout the body via the blood and is transported into the cells as normal glucose. Inside the cell, the FDG is only metabolised one step into FDG-6-P and stops (Pacák, Točík, and Černý, 1969; Weinberg, 2006), whereas regular glucose is further metabolised to create ATP (cell energy) with CO₂ and H₂O as rest

products. Therefore, the quantity of18F in a tissue is approximately proportional to the metabolic rate of glucose in the cells of a tissue. Thus, FDG can be used to map regions with high glucose metabolism and high glucose uptake, such as tumours that metabolise through glycolysis (Warburg, Wind, and Negelein, 1927; Vander Heiden, Cantley, and Thompson, 2009).

Some isotopes can directly be used to examine organs or tissue without being at-tached to a tracer molecule. E.g. 124I, which is used for examination of the thyroid’s functionality. Iodine has a biological function in itself and is distributed in the thyroid proportionally to the tissue metabolism.

2.4.2 Detectors

Detectors for ionizing radiation convert incoming photons into electrical signals. In the case of PET, so-called scintillation crystals provide a first step, where the 511 keV photons interact with the crystal and give rise to the emission of visible light (blue ∼ 400 nm). The light is, in turn, converted to an electrical signal using either a photocathode with a Photo-Multiplier (PM) tube or a Silicon Photo-Multiplier (SiPM) (Knoll, 1980). The generated electrical signal has an amplitude that is proportional to the energy deposited in the crystal, which is one of the components enabling the discrimination of pulses in a selected energy interval. Second discriminator is the timing discrimination called the coincidence window (Wernick and Aarsvold, 2004). For the PET/CT used in this thesis photons with energy below 425 keV and above 650 keV, and detector events separated with more than 4.9 ns are discriminated.

To ensure photon interaction of the relatively high energy photons in PET, the detector elements are made of relatively thick high density scintillating crystals (2-3 cm) made of e.g. LSO (Lutetium Oxyorthosilicate (Lu₂SiO₅(Ce))) or LySO (Lutetium Yttrium Oxyorthosilicate), which have high linear attenuation coefficient (Wernick and Aarsvold, 2004). Detector efficiency

2.4.3 Spatial Resolution

The spatial resolution is limited of several factors, e.g. co-linearity (uncertainties in the emission angle of the annihilation photons), scatter of photons, (both can be reduced by decreasing the detector ring diameter), the energy of the emitted β+-particle, and the so-called Depth-of-Interaction (DOI). All factors affecting the spatial resolution can be condensed into a single factor called ultimate spatial resolution (Γ) (Moses, 2011).

Γ = s _d 2 2 + s2_{+ (0.0044R)}2_{. (2.9)}

Here, d is the crystal width, s is the positron range2, and R is the detector ring radius. The result of eq. (2.9) yields the FWHM (the Full width at Half Maximum) of the 2

Positron range is the average radial distance, from the decay site, an emitted positron travels before depositing enough energy to annihilate. Increased positron emission energy increases positron range and degrades the spatial resolution in the PET image.

point-spread function in units of [m].

The DOI effect arise due to, the arrangement of the detector elements in the PET camera, and the lack of collimators. The photons originating from the ring centre hit all detector elements perpendicularly and interact in one detector crystal. The spatial resolution will be defined primarily by the crystal width d of the detector elements. Photons originating off-centre will have a larger apparent detector-crystal width d0 and may interact with more than one detector crystals, increasing the uncertainty in the radio-tracer position. This is illustrated in figure 2.7 and described mathematically as:

d0 = d cos θ + x sin θ (2.10)

Here, x is the depth of the detector and θ is the angle of incidence on the detector element (Cherry, Sorenson, and Phelps, 2012).

In the detector-ring centre the effective detector-crystal width is smallest and it increases moving radially outward, toward the detector ring circumference. As a conse-quence, the resulting spatial resolution in a PET image will be the highest in the image centre. d’ x d LOR A 7 I so

:

:

Figure 2.7: Illustration of the variation of the apparent detector width due to off-centre location of the

LOR.

2.4.4 Time-Of-Flight TOF

In theory, if the exact time of each coincidence-photon detection is known, the time difference can be used to deduce the decay site to a point along the LOR, see figure 2.8. Current technology offers timing of coincidence photons but within a time window of 300-600 ps and is called Time-Of-Flight (TOF) (Vandenberghe et al., 2016). TOF increases the Signal-to-Noise Ratio (SNR) of the reconstructed image. The image SNR is improved approximately as SNR gain ∝ D/∆x, where D is the object size and ∆x

is the off-axis distance of the detector ring (Budinger, 1983). The increase in SNR can either be used to; shorten the acquisition time, or reduce the activity injected.

Without the TOF, the probability of the annihilation-event origin is assumed to be equal along the LOR or alternatively, if the object border is known, the probability is 0 outside the object boarder and equal along the LOR inside the object surface. The TOF measurement further decreases the probability along the LOR.

Full body PET detector-ring diameter, depending on modality, ranges between 60 cm to 80 cm. The PET/CT used in this thesis has a diameter of 70 cm and a photon traverses this distance in approximately 2.3 ns. For annihilation photons that originates within this diameter, the coincidence photons will hit each respective detector within the time window given by eq. (2.11).

∆t = 2∆x

c ⇒ ∆x =

∆tc

2 . (2.11)

Here, ∆t is the difference in time of detection, ∆x is the difference from the middle of the LOR from which the annihilation originated and c is the speed of light. The PET system used in this thesis offer timing information with precision of ∆t = 549 ps (Healthcare, 2018), thus allowing ∆x to be determined within 8.2 cm instead of 70 cm. Therefore, the TOF does not improve the spatial resolution, because it requires ∆x < system spatial resolution.

∆x

T

t

Centre

_Detector

Detector

.it#*im:b.wn

-Figure 2.8: Illustration of annihilation photons, originating off-centre, traversing air. See eq. 2.11

2.4.5 Attenuation and Scatter Correction in PET

As discussed in section 2.1 annihilation photons have a probability to interact with tissue, which results in attenuation and when performing quantitative PET this has to be corrected for.

When the first PET systems were developed, this problem had previously been en-countered in the development of the Single Photon Emission Tomography (SPECT). To date, several approaches to counteract attenuation and scatter effects have been devel-oped, including; radioisotope transmission-, transmissionless algorithm-, and CT-based methods.

Prior to the combined PET/CT, the most established method in ET was the trans-mission rod method described in the introduction. Where the attenuation throughout the body was mapped by rotating an external mono-energetic photon-emitting radioiso-tope on a rod, preferably with a photon energy equal to the radioisoradioiso-tope used in the examination. The intensity of the beam was measured on the opposite side of the body

and compared to the intensity without the body (Zaidi and Hasegawa, 2003). The re-sulting attenuation map was used to quantify attenuation and proportionally amplify the photon signals.

As discussed in section 2.1.1, the main interaction for 511 keV photons in tissue and bone is Compton scattering (>99.7% (Zaidi and Hasegawa, 2003)). Analytically, the attenuation of a narrow photon beam is the integral of the attenuation along its path, see eq. (2.12) (Zaidi and Hasegawa, 2003).

Φ(s) = Φ0e[−

R

sµ(x,y)ds]. (2.12)

Here, Φ is the photon fluence after distance s, Φ₀ is the initial photon fluence, and µ(x, y) is the linear attenuation coefficient at point (x, y) in the object. Attenuation is the probability of a photon being scattered and missing the detector.

In PET where two photons are needed to get a true count, the probability of a true count, is equal to the detector elements efficiency squared (Det2_{ef f}) times the total atten-uation along the LOR i.e. independent of annihilation origin. The resulting attenatten-uation factor is expressed in eq. (2.13).

For an object of a homogeneous attenuating material surrounded by air and placed in-side a PET detector ring with diameter ◦ = S, see figure 2.9, the probability P (Coincidence) of both annihilation photons hitting respective detector is equal to

P (Coincidence) = P (Det1)P (Det2) = e−µ(D−x)e−µx= e−µ(D−x+x)= e−µD. (2.13)

Here, D is the diameter of object that the photons traverse, x is the depth in the object where the annihilation occurred and µ is the linear attenuation coefficient of the object.

D-x

D

x

Detector

_Detector

S

-Figure 2.9: Two annihilation photons traversing a homogeneous material with attenuation = µ

sur-rounded by air. Schematic illustration of eq. 2.13

Photon scatter is difficult to handle compared to attenuation. Approaches to correct for scatter in PET includes; energy window settings, modelling based on measurements of setups with and without scatter, hardware, spectral analysis, convolution and decon-volution, scatter estimation, statistical, and iterative approaches (Zaidi and Montandon, 2007).

Due to the arrangement (geometry) of the detectors, the probability of scattered coincidences is largest over the detector-ring centre (Sánchez-Crespo and Larsson, 2006) and, if uncorrected, appears in the reconstructed PET image as background blurring that intensifies towards the image centre (Cherry, Sorenson, and Phelps, 2012). Figure 2.10 illustrates qualitatively the effects on the line profile through an object centre with a homogeneous activity concentration (CA) when incorporating AC and SC (note that

noise is neglected). Because scatter coincidences yield counts to the acquired PET

Measured without

corrections AttenuationCorrected Scatter CorrectedAttenuation and

True Profile Object with Homogeneous C C line profile through centre [Bq/ml] Line profile A A

Figure 2.10: Qualitative illustration of the impact of AC and SC for quantitative PET imaging of an

object with a homogeneous Activity Concentration (CA) (without noise).

image, the order in which the AC and SC are implemented make both a quantitative and a qualitative difference. If the AC is applied before the SC, the scatter coincidences are amplified, and SC becomes less effective. Therefore, the SC is always applied prior to the AC to the reconstruction (Cherry, Sorenson, and Phelps, 2012). In this study, the SC was applied prior to the AC.

2.5 Using CT Data in PET Analyses

2.5.1 Attenuation Correction CT (ACCT)

Modern clinical PET modalities (except for PET/MR) are equipped with a CT in ad-dition to the ring of PET detectors. The CT can be used for diagnostic purposes, i.e. for producing CT anatomical images with diagnostic quality, but also for AC and SC, called Attenuation Correction CT (ACCT).

An ACCT can be described as a simplified anatomy image, which gives information about position and delineation of the organs, with sufficient data to achieve correct Hounsfield Unit (HU)3 in respective organ. The ACCT is translated from HU into 3_{Hounsfield Unit (HU) is the unit of CT pixel values, which corresponds to the electron density/linear}

attenuation in the pixel/voxel. HU ranges from -1000 HU for air, to 0 HU for water, and > 0 HU for materials with higher electron density than water and increases approximately linearly with increased

linear attenuation coefficient values for 511 keV photons with a bilinear curve see figure 2.11 (John R. Votaw, 2015).

Two reasons make high spatial resolution redundant in an ACCT: the image spatial resolution is down sampled to match PET, and the image is filtered with a wide Gaussian-filter to remove noise, hence, removes details. An example of an ACCT image, of the NEMA Body phantom with the PET camera used in this thesis, that is converted from HU to linear attenuation coefficients is presented in figure 2.12.

Figure 2.11: Bilinear HU-to-µ511keVconversion plot. (John R. Votaw, 2015)

50 100 150 200 250 300 350 400 450 500 50 100 150 200 250 300 350 400 450 500 -200 -150 -100 -50 0 50 100 150 200 [HU]

(a) 512×512 CT image of phantom.

20 40 60 80 100 120 20 40 60 80 100 120 0 0.001 0.002 0.003 0.004 0.005 0.006 0.007 0.008 0.009 0.01 [1/cm]

(b) 128×128 Attenuation map for 511 keV pho-tons.

Figure 2.12: Example of ACCT to attenuation map conversion: 140 kVp and 775 mAseff ACCT (a)

converted to an attenuation map (b).

2.5.2 Iterative Reconstruction

The acquired PET data is usually reconstructed with an iterative method, instead of FBP, such as MLEM (Maximum Likelihood Estimation Maximization) (Shepp and Vardi, 1982) or OSEM (Ordered Subset Estimation Maximization) (Hudson and Larkin, 1994). Both methods follow a similar algorithm; Creating a guess image and forward projected it, arbitrarily compare each projection with the acquired projections and cor-rect the guess projection, back project and update the guess image. This is repeated until the guess image has converged to the wanted image (Zaidi and Hasegawa, 2003). A detailed flowchart of the MLEM algorithm is found in figure 2.13.

The SC and AC based on the ACCT data can be implemented into the algorithm into the so-called system matrix, which is used to forward project (convert to a sinogram) the guess image in each iteration of the MLEM algorithm.

The MLEM method is a computer-power heavy algorithm i.e. slow. The OSEM method converges faster than MLEM, but at the cost of image quality, by splitting the projections into subsets and performs updates on the guess image after comparing and updating every projection in a subset. One iteration is defined as: all projections have been compared and updated. The OSEM algorithm performs MLEM algorithm once per subset during one iteration.

AC map Guess Forward projecting Guess Measured projection data Comparison Step Forward Project and Correct Guess Iteration stop or continue? Final Image Update Guess Forward projection of

Guess includes SC and AC based on the ACCT

ammo

¥

I

2.5.3 Quantitation Achieved Consistently – Q.AC.

A white paper from GE Healthcare published in 2012 describes a new ACCT recon-struction algorithm called Quantitation Achieved Consistently (Q.AC.). This allows for correct HU values with ACCT using low mAseff, and kVp settings corresponding to a

significantly lower dose than previously possible (Lonn, 2012). The Q.AC. is developed specifically to produce 511 keV photon attenuation maps for AC and SC.

The Q.AC. is presented to yield similar line profiles using 5 mAseff with 100 kVp as

compared to acquisitions using 50 mAs_eff with 100 kVp on a homogeneous oval water-phantom. The mechanisms of the Q.AC are not described, but it is specified to be incorporated into the sinogram prior to the image being reconstructed.

It suggests that the application of this algorithm in PET examinations would enable a possibility to reduce ACCT dose while preserving the quantitative data.

In figure 2.14, CT images acquired with 2.3 mAseff and 80 kVp using regular soft CT

algorithm compared to the Q.AC algorithm and respective 511 keV attenuation map are shown.

-200 -150 -100 -50 0 50 100 150 200 -200 -150 -100 -50 0 50 100 150 200 0 0.001 0.002 0.003 0.004 0.005 0.006 0.007 0.008 0.009 0.01 0 0.001 0.002 0.003 0.004 0.005 0.006 0.007 0.008 0.009 0.01

a)

b)

c)

d)

Figure 2.14: CT acquisition with 80 kVp and 2.3 mAseff. CT reconstructions using, a) regular soft CT

algorithm, and b) the Q.AC. algorithm. Pixels units [HU].

Attenuation maps µ511keV based on CT reconstruction using, c) regular soft CT algorithm, and d) the

3

Material and Methods

PET- and CT-images were acquired with an GE PET/CT Discovery MI DR modality, see figure 3.1.

In the list below, the procedure for one set of PET data acquisition is summarized in six steps.

A detailed protocol for the phantom preparation and acquisition is found in appendix C.

Figure 3.1: Photo of the PET/CT Discovery MI DR used for acquisition

1. Prepare the phantom with the wanted C_A in each phantom compartment inside the hotlab.

2. Align the phantom with guiding lasers.

3. Perform 1 bed, 60 min PET acquisition in list mode.

4. Perform one CT acquisition for each of mAs_eff listed in table 3.1 in combination with [80, 100, 120, 140] kVp for CT reconstruction algorithm: Q.AC. and regular soft CT.

5. Perform one PET reconstruction, for every acquired CT with acquisition time T = 276 s, with OSEM: x subsets and y iterations.

6. Export PET DICOM data and analyse. 3.2 Phantoms

Measurements were performed on two different sized phantoms; The NEMA body phan-tom that has several compartments that can be filled with activity and it is dimensioned; laterally 30 cm, and 23 Anterior-Posterior (AP). The large volume, referred to as the background volume is approximately 10 L. Six different sized spheres of diameter ◦∈[10, 13, 17, 22, 28, 37] mm, and a cylindrical lung-insert situated at the phantom centre which contain air and low-attenuating Styrofoam. In this study, this phantom represents pae-diatric and small/thin adult patients (see figure 3.2).

The second phantom is the NEMA body phantom with an additional extension ring. The extension ring is a custom made PMMA-plastic extension that adds laterally 10 cm per side and 2 cm AP per side (see figure 3.3). This phantom represents large sized patients.

The NEMA body phantom compartments are to be filled with different Activity Concentrations (CA).

(a) Photo

(b) Schematic illustration of phantom and ROI drawing.

Figure 3.3: Photo of the extension ring made of PMMA (dimensions: 50 cm laterally and 27 cm AP).

It surrounds the NEMA body phantom to increase the attenuation and this simulated a medium/large sized patient.

3.3 NEMA NU 2-2012 – Performance Measurement of PET Systems The National Electrical Manufacturers Association (NEMA) NU protocol is an inter-national standard for making performance measurements e.g. Quality Assurance (QA) and Quality Control (QC) of PET modalities. The protocol includes five tests, of which this thesis only applies the test on “Image quality, accuracy of attenuation correction and scatter correction”.

This test is based on measurements with the NEMA body phantom, see section 3.2. The four smaller spheres are filled with the highest Activity Concentration (C_A) (hot spheres), the background volume is filled with 1/4 of the C_Ain the hot spheres and the two largest spheres are filled with water without activity. In this work, the four smallest spheres are filled with an C_A of approximately 20 kBq/ml of 18F at the start of the acquisition.

The NEMA NU-2 2012 protocol specifies where all the Region of Interests (ROIs) are to be drawn and their respective sizes. In subfigure 3.2b an illustration from the NEMA NU-2 2012 protocol is presented (NEMA, 2013) where the ROIs for the hot (black), and cold (white) spheres, the lung insert and a suggestion of where to draw background ROIs are seen. As illustrated in 3.2b, there is one ROI per sphere size drawn in the slice where the maximum sphere diameter is found. For each sphere size there are also 12 background ROIs per slice in five slices and there is also one (◦ = 30 ± 2 mm) ROI per slice in the centre of the cylindrical lung insert within 10 mm of its axial edges. The ROIs are used to determine the following quantities in the recorded PET images;

• ∆lung: Accuracy in AC and SC

∆lung is the relative difference between the measured CA in the lung insert ROIs

and the 37 mm background ROIs. There is one circular ROI of diameter 30 mm ± 2 mm drawn in the centre of the lung insert in every slice within 10 mm from

each axial edge of the phantom. ∆lung =

Clung

CB,37mm

· 100%. (3.1)

Here, C_lung is the average of the average counts in the lung ROIs and CB,37 mm

is the average of the average counts in the 60 background spheres of diameter 37 mm. True value of ∆lung is 0 %, hence, low percentage of ∆lung suggests high

image quality.

• QC: Analysis of Cold Spheres

The two larger spheres without activity represent cold spots in a PET image and may suffer from spill-in effects because of the hotter background, it is also called the Partial Volume Effect (PVE). The QC is determined with eq. (3.2).

QC,i = 1 −

CC,i

CB,i

!

· 100%. (3.2)

Here, CC,i is the measured counts in a circular ROI with a diameter equal to

respective cold sphere diameter and CB,i is the measured average counts of all 60

background ROIs of equal size to the respective cold sphere. True contrast is 100 % hence, high percentage of QC suggests high image quality.

• Q_H: Analysis of hot spheres

The four smallest spheres filled with the highest CA, are analysed with eq. (3.3).

QH,i=   CH,i CB,i − 1 aH aB − 1  · 100%. (3.3)

Here, C_H,i is the measured counts in a circular ROI with a diameter equal to respective hot sphere diameter, CB,iis the measured average counts of all 60

back-ground ROIs of equal size to respective hot sphere, aH is the measured CA in the

hot spheres and a_B is the measured C_A in the background volume. True contrast is 100 % hence, high percentage of QH suggests high image quality.

• N : Background Variability

The background variability quantifies the variation of noise in the background. There are in total 60 ROIs of each sphere size positioned in the background of 5 slices and for each sphere size, the background variability N_i is quantified as:

Ni=

SDi

CB,i

· 100%. (3.4)

Here, SDi is the standard deviation of the average counts of the 60 background

ROIs of the specific sphere size and C_B,i is the average counts of the respective 60 ROIs. Low percentage of N suggests high image quality.

3.4 Instrument Settings (Measurement Data collection)

3.4.1 ACCT Acquisition Settings

To determine which range of mAs_eff that should be investigated, an anthropomorphic water phantom was used to simulate clinically representative attenuation. The HU in the phantom centre was measured to estimate the mAseff at which the reconstruction

algorithms failed to achieve correct HU value for water, see figure 3.4. The finale values of mAs_eff are listed in table 3.1.

Figure 3.4: Anthropomorphic phantom filled with water and cylindrical spine insert. Two legs were

positioned beside the phantom to simulate clinically relevant attenuation.

Acquisition # 1 2 3 4 5 6 7 8 mA 10 20 30 40 50 60 70 80 trot 0.35 0.35 0.35 0.35 0.35 0.35 0.35 0.35 mAseff 2.3 4.6 6.9 9.1 11 14 16 18 Acquisition # 9 10 11 12 13 14 15 mA 90 130 180 250 380 400 400 trot 0.35 0.35 0.35 0.35 0.35 0.5 1 mAs_eff 21 30 41 57 87 130 260

Table 3.1: List of the different mAseff to investigate in combination concluded from the

3.4.2 PET Acquisition Settings

The NEMA protocol does not specify a data collection time, instead eq. (3.5) is given to determine the acquisition time T depending on the specifications of the PET modality (given CA=20 kBq/ml).

T = 30min

100cm · axial step. (3.5)

Here, axial step is the length of one bed position i.e. the number of PET slices multiplied with the width of one detector element (along z-axis). For GE PET/CT Discovery MI DR the axial step = 15.4 cm and given by eq. (3.5) the sample time is T = 4.6 min. This sample time was acquired by selecting the first 4.6 min of the 60 min PET acquisition which was possible due to acquisition being performed in list mode.

3.5 Data Analysis

To analyse large number of images, a MATLAB script was written to automate the NEMA NU-2012 analysis described in section 3.3. Mapping of the spheres, the cylin-drical lung-insert, and the background-ROIs positions were determined in an image and corresponding masks (matrix containing zeros and ones mapping ROIs) were made to extract the wanted data from the images. See figure 3.5.

Several PET acquisitions were performed and with guiding lasers and markings the phantom was positioned consistently between acquisitions. However, due to human factors, the hot and cold spheres in all PET images were shifted compared to the pre-drawn ROIs. Therefore, the images were resliced and rescaled with the image analysis tool FSL (FSL 2019) functions to match the script template, see figure 3.6.

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

(a) One ROI per sphere size + one ROI for the cylindrical lung insert in the centre.

0 1 2 3 4 5 6

(b) 12 background ROIs per sphere size seen in subfigure (a) and their positioning in the phantom.

-100.0 -80.0 -60.0 -40.0 -20.0 0.0 20.0 40.0 60.0 80.0 100.0 [u n d e fi n e d ]

(a) CT acquisition used as template for data analysis. [HU]

sAT_190220-0474-00001-000001.nii <frame 1> 0.0 1000.0 2000.0 3000.0 4000.0 5000.0 6000.0 7000.0 8000.0 9000.0 10000.0 [undefined]

(b) Acquired PET fused with CT template image (a) to highlight the erroneous positioning of the phantom prior to acquisition and the resulting shift of the hot and cold spheres. [Arbitrary unit]_{zz_imreslizzedsAT_190220-0474-00001-000001.nii <frame 1>} 0.0 1000.0 2000.0 3000.0 4000.0 5000.0 6000.0 7000.0 8000.0 9000.0 10000.0 [undefined]

(c) PET image seen in (b), resliced and rescaled to match the CT template, fused with CT template to highlight the hot and cold spheres are now correctly positioned. [Arbitrary unit]

4

Results

The main purpose of this work was to examine whether the patient doses from ACCT data collection may be lowered in PET/CT examinations, in particular in the light of the newly developed Q.AC. CT reconstruction algorithm for ACCT. As means for the evaluation, the NEMA NU 2-2012 protocol described in section 3.3 was followed, focusing on the results in terms of retained PET-image quantitative quality.

Accordingly, ACCT settings were varied, and CT reconstructions were made utiliz-ing the new Q.AC. algorithm and the regular soft CT algorithm. All PET images were recorded, reconstructed and analysed as described in section 3.4 and 3.5. In this section, the resulting NEMA NU 2-2012 quality parameters obtained for the varies ACCT set-tings are presented. Error bars of ±σ are included, where the uncertainties have been calculated according to Appendix B.

4.1 Measured Activity in Lung Insert ∆lung

In figure 4.1, the ∆lung versus mAseff measurement results for a set of kVp settings

on performed on the NEMA body phantom are presented. Subfigure titled Q.AC. in 4.1 displays PET data reconstruction results of based on Q.AC. reconstructed ACCT. Subfigure titled regular soft CT algorithm in 4.1, displays PET data reconstruction results based on the regular soft CT reconstructed ACCT. The results are representative for paediatric patients and small/thin adults.

Figure 4.2 displays the corresponding plots for the NEMA body phantom with ex-tension ring, where the results are representative for mid- to large-sized patients.

2.3 4.6 6.9 14 30 58 130 260 mAs eff 2 4 6 8 10 12 14 16 18 20 lung [%] Q.AC 2.3 4.6 6.9 14 30 58 130 260 mAs eff 2 4 6 8 10 12 14 16 18 20 lung [%]

Regular Soft CT Algorithm

80 kVp 100 kVp 120 kVp 140 kVp Figure 4.1: NEMA Bo dy Phan tom Results ∆lung vs. mAs eff .

2.3 4.6 6.9 14 30 58 130 260 mAs eff 5 10 15 20 25 30 35 40 45 50 55 lung [%] Q.AC 2.3 4.6 6.9 14 30 58 130 260 mAs eff 5 10 15 20 25 30 35 40 45 50 55 _lung [%]

Regular Soft CT Algorithm

80 kVp 100 kVp 120 kVp 140 kVp Figure 4.2: NEMA Bo dy Phan tom with Extension Ring Results ∆ lung vs. mAs eff .

4.2 Measured Contrast in Cold Spheres QC

In the figures of this section, the results for the measurements of the Q_C for the 37 mm and 28 mm sphere are displayed for various kVp and mAs_eff combinations. In each sub-figure, two lines per sphere size are found; the results from using the regular soft CT reconstruction algorithm, and the results using the Q.AC. CT reconstruction algorithm. Figure 4.3 and 4.4 displays measurements of the NEMA body phantom and NEMA body phantom with extension ring respectively.

2.3 4.6 6.9 14 30 58 130 260 mAs eff 80 82 84.4 86.3 88 90 92 94 96 Q C [%] 80 kVp 2.3 4.6 6.9 14 30 58 130 260 mAs eff 80 82 84.3 86.3 88 90 92 94 96 Q C [%] 100 kVp 2.3 4.6 6.9 14 30 58 130 260 mAs eff 80 82 84.3 86.3 88 90 92 94 96 Q C [%] 120 kVp 2.3 4.6 6.9 14 30 58 130 260 mAs eff 80 82 84.2 86.2 88 90 92 94 96 Q C [%] 140 kVp

28 mm Q.AC. 28 mm Regular soft CT algorithm 37 mm Q.AC. 37 mm Regular soft CT algorithm Contrast with max mAs

eff Figure 4.3: NEMA Bo dy Phan tom QC vs. mAs eff Results: 37 mm sphere comparing Q.A C. to Regular soft CT algorithm.

2.3 4.6 6.9 14 30 58 130 260 mAs eff 57 60 64 70 74.1 80 82 Q C [%] 80 kVp 2.3 4.6 6.9 14 30 58 130 260 mAs eff 57 60 63.3 70 74.3 80 82 Q C [%] 100 kVp 2.3 4.6 6.9 14 30 58 130 260 mAs eff 57 60 63.4 70 74.4 80 82 Q C [%] 120 kVp 2.3 4.6 6.9 14 30 58 130 260 mAs eff 57 60 63.7 70 74.6 80 82 Q C [%] 140 kVp

28 mm Q.AC. 28 mm Regular soft CT algorithm 37 mm Q.AC. 37 mm Regular soft CT algorithm Contrast with max mAs

eff Figure 4.4: NEMA Bo dy Phan tom with extension ring QC vs. mAs eff Results: 37 mm sphere comparing Q.A C . to Regular soft CT algorithm.

4.3 Measured Contrast in Hot Spheres QH

In the figures of this section, the results of the Q_H for the smallest (10 mm) and the largest (22 mm) hot spheres are displayed for various kVp and mAs_eff combinations. In each sub-figure, two lines per sphere size are found; the results from using the regular soft CT reconstruction algorithm, and the results using the Q.AC. CT reconstruction algorithm. Figure 4.5 and 4.6 displays measurements of the NEMA body phantom and NEMA body phantom with extension ring respectively. As mentioned in section 3.3, high percentage indicates better image contrast.