Update on novel trends in PET/CT technology and its clinical applications

Abstract

After a brief history of the major evolutions of positron emission tomography since its introduction in 1972, this article reviews the recent improvements and novel trends in positron emission tomography with a special focus on the time of flight that is currently the major research topic. Novel emerging acquisition modalities, such as dual tracer acquisition, inline hadron therapy dose imaging and yttrium-90 imaging are reviewed.

INTRODUCTION

Although single-photon emission CT (SPECT) has not changed much since its beginning [most SPECT are still made of sodium iodide (NaI) photomultiplier tube (PMT) detectors as invented by Anger¹ in 1958 and equipped with a parallel hole collimator], positron emission tomography (PET) has seen several major improvements since the proof of concept made by Burnham and Brownell² in 1972. This first system used a rotating dual-head detector made of an array of 127 NaI crystals, each coupled to a few PMTs in order to reduce the dead time.

Imaging radioactivity requires the knowledge of the incoming direction of the recorded gamma rays. Unlike SPECT where a mechanical collimator is used to discard the gamma rays not coming perpendicularly to the crystal (Figure 1a), in PET, the incoming direction of the two annihilation photons detected in coincidence is given by the straight line connecting the two hit crystals, also called the line of response (LOR) (Figure 1b). This electronic collimation has the major benefit to allow the exploitation of all detected coincidences, resulting in a final sensitivity approximately two orders of magnitude higher than that of SPECT. Moreover, this electronic collimation ensures a spatial resolution almost uniform in the field of view (FOV), whereas it significantly worsens in SPECT when moving away from the collimator (Figure 1).

Figure 1.

(A,B) Unscaled transverse section illustrating the spatial resolution and sensitivity improvement obtained in positron emission tomography (PET) by the replacement of the mechanical collimator by an electronic collimation. Sensitivity: in single-photon emission CT (SPECT), γ-rays going from the decay position c to the crystal position a are ruled out by the collimator, while they are detected in PET. Spatial resolution: Lines 1 and 2 show the accuracy limits on the decay position b when the gamma ray is detected at the crystal position a, note that in SPECT, this accuracy worsens when moving away from the collimator while it is almost uniform in PET. Time of flight (TOF)-PET still goes a step beyond by giving the probability of the decay location along the line of response.

Ter-Pogossian et al³ in 1975 boosted the efficiency and reduced the random coincidences rate by using a hexagonal array of NaI-PMT detectors surrounding the patient. Random coincidences result from nearly simultaneous detections of two 511-keV photons but originating from two different annihilations while the two reminder photons are missed by the system (random coincidences involving more than two detected photons are easily discarded by the system). Their rate is thus inversely proportional to the system efficiency. Obviously, random coincidences provide wrong information about the annihilation direction and therefore reduce the contrast in the reconstructed image.

The breakthrough step in this first decade was the introduction by Thompson et al⁴ in 1979 of bismuth germanium oxide (BGO) crystals used in a full-ring detector setup. Although BGO suffers from light yields lower than that of NaI, worsening the energy resolution, its higher efficiency combined with the full-ring detector significantly reduces the random to true coincidences ratio down to a value compatible with human studies. This ring setup still remains the design of all modern PETs.

Early BGO PETs had a short axial FOV, but thanks to the reduction of the BGO production cost, it was continuously extended by stacking several rings of pixilated BGO blocks to reach about 20 cm in the nineties. Owing to the poor energy resolution of BGO, the axial FOV extension induced an increased fraction of detected scattered coincidences. This problem was overcome by adding a mechanical axial collimation made of thin tungsten rings called septa. With this setup, the PET behaves as several independent small axial FOV PETs, all simultaneously imaging different slices of the patients. Obviously, this two-dimensional acquisition mode reduces the system efficiency by discarding axially tilted LORs. A motorized retraction of the septa was incorporated to allow three-dimensional acquisitions, mainly dedicated to brain acquisitions for which the scattered coincidences fraction was low due to the small diameter of the head.

Up to that time PET was mainly a research tool using different radiotracers: oxygen-15 (¹⁵O)-labelled water or nitrogen-13-labelled ammonia to measure blood flow in organs; ¹⁵O-labelled oxygen to measure oxygen metabolism; and fluorine-18 fludeoxyglucose (¹⁸F-FDG) to monitor glucose metabolism. The half-life and positron energy of fluorine-18 (¹⁸F) are nearly optimal for labelling process, radiotracers delivery, patient radioprotection and positron imaging. Already in the eighties, it was shown that the tumours avidly take ¹⁸F-FDG up, and it became clear in the nineties that ¹⁸F-FDG-PET was an interesting tool in the diagnosis, management and treatment of cancers. All these initiated the ¹⁸F-FDG-PET reimbursement and the clinical use of PET imaging. Nowadays, ¹⁸F-FDG imaging in cancer still constitutes 92% of the PET routine vs 6% and 2% for cardiology and neurology, respectively.⁵

Besides all these interesting features for cancer applications, ¹⁸F-FDG has the drawback to be also taken up by normal tissues such as the brain, liver, and heart and inflammatory tissues and to also accumulate in the urinary bladder. These unwanted uptakes hamper the visualization of tumours and their identification motivating further developments in PET imaging.

Thanks to faster and faster personal computers, PET imaging started to continuously profit from reconstruction algorithm improvements that better and better compensate for the noise and scatter fraction present in the acquisition.^6,7 Indeed, unlike in CT where the patients are irradiated only during the scan, most of the gamma rays are emitted before and after the imaging acquisition in nuclear medicine. This explains why noise issues still exist in nuclear medicine imaging for doses delivered to the patient similar to the ones in CT.

Combined PET/CT systems were commercially launched in 2001.⁹ CT has two purposes: allowing the localization and identification of active regions on an accurate anatomical image and generating an attenuation map used to compensate gamma ray attenuation in PET imaging. These two CT applications do not require a tissue contrast as high as that needed for diagnostics purpose. Therefore, CT scans combined with PET are usually performed with low beam intensity. CT also benefited this past decade from major improvements well known by BJR readers.

2001 also saw the real starting of whole-body three-dimensional acquisitions thanks to the introduction of gadolinium orthosilicate (GSO) and lutetium orthosilicate (LSO): two crystals faster and brighter than BGO, in that way reducing dead time, and random and scatter coincidences fraction.¹⁰ GSO was quickly abandoned to the benefit of lutetium-based crystals, i.e. LSO and lutetium–yttrium orthosilicate (LYSO), whose timing resolution allowed the introduction of time of flight (TOF) PET in 2005.¹¹ In this system, the measure of the delay separating the two 511-keV photon detection provides additional information on the annihilation location (Figure 1b). In this review, we will see that since its introduction, the timing resolution of TOF-PET systems has continuously been improved due to an intense quest for the holy grail, i.e. achieving a PET system directly recording the annihilation location, decay by decay. While tending to this holy grail, TOF-PET has not only improved standard routine ¹⁸F-FDG imaging but has also boosted the development of novel acquisition modes, such as in-line proton therapy dose imaging and yttrium-90 (⁹⁰Y) imaging. This last one is becoming the standard in liver radioembolization performed by the interventional radiologist.

TIME-OF-FLIGHT POSITRON EMISSION TOMOGRAPHY IMPROVEMENTS

Late improvements

In 2005, a breakthrough was initiated in PET with the first successful commercial launching of a TOF-PET system.¹¹ Such attempts already occurred in the eighties, but failed due to the low sensitivity of fast crystal available at that time.^12,13 TOF-PET consists in measuring the delay Δt in-between the detections of the two annihilation 511-keV photons (Figure 1b). TOF information directly provides the annihilation location on the LOR by a simple multiplication:

$$r=\hspace{0.17em}\frac{c}{2}\Delta t\hspace{0.17em}$$ (1)

where c is the speed of light in vacuum, i.e. 2.99792458 × 10⁸ m s⁻¹, and r is the offset from the LOR centre.

The breakthrough nature of TOF-PET invention is related to the fact that PET image reconstruction, as many tomographic modalities and deconvolution processes, is an ill-posed, or ill-conditioned, problem: a small perturbation in the measured data can induce large and intense unpredictable artefacts in the reconstructed image^14,15 (for a more wide public explanation of the ill-posed nature of PET (Walrand et al¹⁵). Let us recall that gamma counting is always affected by a Poisson noise inherent to the decay process, independently of the counting technique.

The activity in a voxel is reconstructed from all the LORs crossing this voxel, thus globally from the whole slice activity. As a result, correlated and structured noise is present in the reconstructed transverse slices, i.e. the noise present in one voxel depends on the values of all the other voxels in the slice (Figure 2). This induces streak artefacts in the vicinity of high active tissues hampering the visualization and quantification of the neighbouring tissues.¹⁵ By contrast, a small error ε in the measurement of Δt just induces the predictable error cε/2 on the annihilation location.

Figure 2.

Reduction of the streak artefacts (pointed by the arrows) present in the vicinity of the bladder in non-time of flight (TOF) reconstruction. Courtesy of Dr Joel Karp.

When TOF-PET systems will be able to measure Δt with an error <25 ps, the annihilation position will be known with an accuracy better than 3.75 mm allowing to directly locate in which 4-mm-size voxel the annihilation occurred. From there, the reconstruction process will no longer be needed and the computer screen will display the events added one by one in the transverse slices during the acquisition, such as in nuclear medicine planar acquisition. As a result, the streak artefacts will vanish, and the noise in the transverse slice will no longer be correlated and will follow the Poisson distribution as in planar imaging.

This PET revolution is on its tracks: within one decade, the initial TOF resolution has been improved by a factor of 2, moving from 650 ps at the launching time down to 320 ps in new digital TOF-PET systems.⁶ These TOF resolutions still result in an indetermination of 9 and 4.5 cm on the annihilation location, respectively, i.e. still one order of magnitude higher than the ultimate goal. However, this partial information already improves the reconstructed image quality, especially in low count acquisitions, such as in corpulent patients where the number of detected gamma rays is significantly reduced by attenuation.

Indeed, all reconstruction algorithms, analytical or iterative, contain a back-projection step in which detected counts in conventional PET are uniformly summed on their corresponding LOR. In TOF-PET, counts are summed only along a small LOR segment centred on the most probable annihilation location (Figure 1b). As a result, the noise correlation length is reduced and consequently so are the size and intensity of the streak artefacts (Figure 2). In other words, TOF knowledge improves the problem conditioning.

During three decades, TOF resolution of numerous thin crystal-photosensor tandems have been intensively investigated in two-arm setup experiments (Figure 3). TOF resolution of 75 ps has been reached in this simple setup using thin pure caesium bromide crystal.

Figure 3.

Evolution of time of flight (TOF) resolutions experimentally observed. Straight lines correspond to TOF-positron emission tomography (PET) systems. Symbols correspond to two arms set up using various crystal types [blue: Barium fluoride (BaF2), red: lutetium orthosilicate (LSO) or lutetium–yttrium orthosilicate (LYSO), green: LaBr3, pink: caesium bromide (CsBr), violet: lutetium gadolinium oxyorthosilicate (LGSO), black: liquid Xenon (Xe), Cesium fluoride (CSF) or Aluminium Copper Iron (AlCuFe)] and coupled with various photosensors [open symbols: photomultiplier tube (PMT), solid symbols: avalanche photo diode (APD) or silicon photomultiplier (SiPM)]. Better coincidence timing resolutions for the same crystal–photosensor tandem correspond to thinner crystal (more details and references are given in Walrand and Jamar¹⁷).

The year 2012 showed a shift from experimental TOF investigations to theoretical works supported by Monte Carlo (MC) simulations in order to understand and predict timing resolution of crystal-photosensor tandems.^18–20 Degradation of TOF resolution mainly occurs in the scintillation crystal and for a smaller part in the photosensor. Scintillation is characterized by the rise time τ_r and the decay time τ_d of the light pulse and by the number of scintillation photons N_pe produced by the photoelectric effect. The intrinsic timing resolution R of a thin, fast and light scintillator is:

$$R=2.18\hspace{0.17em}\sqrt{\frac{\hspace{0.17em}{\tau }_{\text{r}}\hspace{0.17em}{\tau }_{\text{d}}}{N_{\text{pe}}}}$$ (2)

Equation (2) is valid if τ_d ≪ τ_r × N_pe, which is the case for LSO : caesium (Ce) [τ_r = 0.07 ns, τ_d = 40 ns, N_pe (511 keV) = 20,500], resulting in an intrinsic TOF resolution of 25 ps.²¹

However, thin crystals are not adapted to PET that requires high detector efficiency Ef, as the true coincidences rate is proportional to Ef² while the random coincidences rate is proportional to Ef. Beside photoelectric and Compton interactions, gamma rays are not affected by the crossed medium and always travel at the speed c. On the contrary, owing to its much larger wavelength, the scintillation light is sensitive to the electric permittivity ε and the magnetic permeability μ of the crossed material and travels in crystals at a speed of c/n_r, where n_r is the refraction index of the crystal (approximately 2 for most crystals). As a result, depending on whether the photoelectric effect occurs in the front or back region of a crystal of thickness H, the arrival time of the scintillation light on the photosensor changes by a delay:

$$\Delta t=\hspace{0.17em}\frac{H}{c}\hspace{0.17em}\left(n_r-1\right)\approx \hspace{0.17em}\frac{H}{c}$$ (3)

As a result, the TOF resolution will be increased by about Δt (approximately 170 ps for H = 25 mm) or equivalently the annihilation location uncertainty will be increased by about the crystal thickness.

Current PET used pixelated crystal, each crystal element being separated from the other ones by a reflective layer. Most of the scintillation photons do not directly reach the photosensor but follow a complex walk resulting from many reflections on the crystal edges (Figure 4). In conventional PET where crystal blocks share light between different PMTs, the first photons reaching the photosensor can come from very different walk length into the crystal (Figure 4a). This feature called photon dispersion still worsens the TOF resolution. In digital PET, a one-to-one coupling is performed between crystal and photosensor pixels. This reduces the walk length variability and thus improves TOF resolution (Figure 4b). GE (Milwaukee, WI) in its digital PET chose to keep the LSO thickness at 25 mm resulting in a TOF resolution of 400 ps,²² whereas Philips (Cleveland, OH) decided to reduce the LYSO thickness down to 19 mm improving the TOF resolution to 320 ps.²³ Determination of the optimal trade-off between sensitivity and TOF resolution is challenging and likely depends on the studies' count rates.

Figure 4.

(a–d) Schematic slice of a crystal (yellow) photosensor block. Reflective coating is represented in grey and scintillation photon in blue. (a) Conventional crystal-4 photomultiplier tubes (PMTs) block. (b) Crystal-silicone photomultiplier (SiPM) block with a one-to-one coupling between crystal and SiPM pixel. (c) Addition of a front SiPM in order to measure the depth of interaction (DOI). (d) Monolithic crystal–SiPM block allowing DOI assessment from the spatiotemporal photon distribution on the SiPM. (e) Schematic drawing of the dichotomous orthogonal symmetric DOI block detector (reprinted from Zhang et al²⁸ with permission from the Institute of Physics Publishing). TOF, time of flight.

Near future improvements

A solution to improve TOF resolution while using thick crystal is to measure the depth of interaction (DOI), i.e. the location depth of the photoelectric effect in the crystal. In that way, as the refraction index of crystals is known with a good accuracy, it is possible to remove the extra delay due to the scintillation light propagation in the crystal from the TOF detected by the photosensors.

An easy, but expensive, way is to add a silicon photomultiplier (SiPM) array in place of the front reflective coating of the crystal block (Figure 4c).¹⁴ Indeed as SiPM²⁴ arrays are very thin, they only slightly attenuate and scatter the 511-keV photons. A DOI resolution of 3 mm has already been obtained by simply using the ratio of photoelectrons number detected by two SiPM arrays.²⁵ Regarding timing resolution, a long crystal becomes virtually equivalent to a 3-mm virtual long crystal but while keeping its high sensitivity. Considering the delay between the arrival times of the scintillation light on the two opposite SiPMs should also allow to still improve the DOI measurement accuracy.

Another way to correct for the scintillation photon walk in the crystal is to use a monolithic crystal coupled to a SiPM array (Figure 4d). In that case, a photoelectric effect occurring in the crystal spreads photons on all the SiPM cells. The photoelectric position can be obtained from the photons' distribution on the SiPM cells and the scintillation photon walk can be estimated by also taking into account the timestamp distribution on the SiPM cells.^26,27 The use of a monolithic crystal has the additional benefits to provide better spatial and energy resolution and to reduce the crystal manufacturing cost. Sub 200-ps TOF resolution and 1-mm spatial resolution using a 20-mm long LSO crystal have been observed.¹⁷ Note that the addition of a second SiPM array on a lateral side of the monolithic crystal should similarly provide a 1-mm DOI resolution, thus better than that obtained with dual-end SiPM arrays.

An elegant method has been proposed to decrease the cost by reducing the number of SiPM cells set at each side of the crystal block.²⁸ Rather than having a reflecting coating on the whole length of the crystal pixel, a small light window is kept between all the crystal pixels: at the top in the x direction and at the bottom along the y direction (Figure 4e). SiPM cells are connected at the top on both x-sides and at the bottom on both y-sides, resulting in four dichotomous N linear cells rather than the two opposite N² arrays. The lighted SiPM cells directly identify the pixel in which the photoelectric effect occurred. The DOI is obtained by comparing the total number of photons detected at the crystal top and bottom. MC simulation reported a DOI resolution of 4.3 mm.

Improvement of crystal speed and brightness is also under investigation in order to decrease the intrinsic timing resolution [Equation (2)]. LaBr₃ : Ce exhibits an intrinsic timing resolution of 20 ps [τ_r = 0.2 ns, τ_d = 17 ns, N_pe (511 keV) = 40,000] but suffers from a lower absorption efficiency than LSO or LYSO. Modifying the doping of LSO is also promising, LSO : Ce,0.4%Ca displayed an intrinsic timing resolution of 15 ps [τ_r = 0.02 ns, τ_d = 33 ns, N_pe (511 keV) = 15,300].^29,30 New generations of lutetium-based crystal such as lutetium gadolinium oxyorthosilicate are also under evaluation (Figure 3).³¹

The fundamental limits imposed by physics are τ_d = 1 ns and N_pe (511 keV) = 100,000.¹⁹ Developed crystals are still far from these limits. Derenzo et al¹⁹ performed extensive MC simulations and found an accurate equation providing the timing resolution as a function of τ_r, τ_d and N_pe (511 keV), of the scintillation dispersion time d in the crystal due to the time walk and of the photosensor photon detection efficiency (PDE) and response time J. In order to better visualize how the different parameters impact the timing resolution, their data can be also approximated by the simple equation:

$$R\approx 2.08\hspace{0.17em}\sqrt{\frac{\hspace{0.17em}\hspace{0.17em}{\tau }_{\text{d}}}{\text{PDE\hspace{0.17em}}\times \hspace{0.17em}{N}_{\text{pe}}}}\hspace{0.17em}\hspace{0.17em}\sqrt{\frac{\hspace{0.17em}{\tau }_{\text{d}}}{\text{PDE}\hspace{0.17em}\times \hspace{0.17em}{N}_{\text{pe}}}+\hspace{0.17em}0.556\hspace{0.17em}\left({\tau }_{\text{r}}+d+J\right)\hspace{0.17em}}$$ (4)

The mean relative deviation of Equation (4) vs the MC simulations is 4% with a maximal |relative deviation| of 11%. For fast and light crystals, such as LSO, the second term in the second square root is predominant, resulting in:

$$R\approx 1.55\hspace{0.17em}\sqrt{\frac{\hspace{0.17em}\hspace{0.17em}{\tau }_{\text{d}}\hspace{0.17em}\hspace{0.17em}\left({\tau }_{\text{r}}+d+J\right)\hspace{0.17em}}{\text{PDE\hspace{0.17em}}\times \hspace{0.17em}{N}_{\text{pe}}}}\hspace{0.17em}\hspace{0.17em}$$ (5)

Derenzo et al¹⁹ showed that a final TOF resolution of 8 ps should be possible using dual end available detectors, provided that a crystal with N_pe (511 keV)/τ_d > 10000 ns⁻¹ could be developed.

All these developments improve PET imaging by reducing the impact of noise. In clinical routine, this can be used for several purposes: to increase the image quality in challenging studies such as tumour visualization in the vicinity of active tissues or in corpulent patients; reduction of the injected activity; or reduction of the acquisition time in order to reduce the study cost while preserving the same imaging quality. Last, it opens PET to novel imaging fields, as illustrated in next sections.

EMERGING ROUTINE ACQUISITION

Yttrium-90 positron emission tomography imaging

⁹⁰Y was considered by the nuclear medicine community as a pure beta emitter up to recently. However, early in 1955, the existence of a very low positron emission was already observed,^32,33 i.e. 32 positron emissions out of 1 million decays.³⁴ The first human ⁹⁰Y-PET imaging was performed in 2009 after a liver radioembolization with resin spheres.³⁵

In post liver radioembolization, thanks to the high ⁹⁰Y activity used (typically 0.8 → 4 GBq), TOF-PET imaging exhibits a good spatial resolution significantly better than that of Bremsstrahlung SPECT^35–37 and is furthermore quantitative using standard manufacturer reconstruction.^38,39 Non-TOF-PET systems can also be used⁴⁰ although with a lower quantification accuracy for LSO-or LYSO-based systems³⁹ and with severe dead time issues for BGO-based systems due to high random count rates originating from the Bremsstrahlung X-rays.⁴¹

Early tumour response assessed by ¹⁸F-FDG-PET proved closely correlated with absorbed dose obtained post radioembolization with ⁹⁰Y TOF-PET imaging.⁴² This supports the implementation of optimized planning. Unfortunately, such implementation is hampered by significant discrepancies between technetium-99m-labelled macroaggregated albumin and actual ⁹⁰Y-microsphere distributions.^43–45 A successful improvement of liver radioembolization by reperforming the procedure on the same day based on the quantitative ⁹⁰Y-PET imaging has already been achieved (Figure 5).⁴⁶ Beside this scarce state-of-the-art treatment, ⁹⁰Y-PET imaging is becoming the standard in post liver radioembolization check^47,48 and has also been recently proposed in radiosynovectomy.⁴⁹

Figure 5.

(a) Axial fluorine-18 fludeoxyglucose (¹⁸F-FDG)-positron emission tomography (PET)/CT performed 3 weeks before yttrium-90 (⁹⁰Y) radioembolization showed large areas of hypermetabolic hepatocellular carcinoma (arrow). (b) Absorbed dose map generated from quantitative ⁹⁰Y-PET/CT after initial infusion of ⁹⁰Y microspheres. Total absorbed doses for the region of interest are shown after the initial infusion and after both infusions. (c) Axial ¹⁸FDG-PET/CT performed 12 weeks after ⁹⁰Y radioembolization showed partial metabolic response to therapy (arrow). Reprinted from Bourgeois et al⁴⁶ with permission from Elsevier Inc.

In ⁹⁰Y peptide receptor radiotherapy (PRRT), activities in tissues are much lower: typically <0.05 GBq in the kidney, which is one of the critical tissues. However, a phantom study showed that ⁹⁰Y-PET imaging was feasible and quantitative for the most taking up tissues, i.e. in tumors and in kidneys.⁵⁰ In this low count rate study, the 650-ps-TOF-PET images were biased by the random coincidences originating from the natural LYSO radioactivity.⁵¹ As a result, the TOF-PET performed less well than an old BGO system.

Recently, a 545-ps-TOF-LYSO-PET system confirmed the quantification feasibility in ⁹⁰Y-PRRT phantoms and in patients, with a spatial resolution clearly surpassing that of Bremsstrahlung SPECT (Figure 6).⁵² Indeed, TOF does not only better condition tomography but also reduces the negative impact of the random coincidences on the signal-to-noise ratio.⁵³ New generations of SiPM-based TOF-PET systems with their TOF resolution reaching 320 ps will still improve the quantification accuracy in ⁹⁰Y-PRRT imaging.^8,16

Figure 6.

Yttrium-90 (⁹⁰Y) positron emission tomography (PET)/CT (first line) and single-photon emission CT (SPECT)/CT (second line) imaging of three patients after administration of peptide receptor radionuclide therapy (PRRT). SPECT/CT and PET/CT scans were acquired at 4 and 6 h post injection, respectively. Patient 1 injected with 1424 MBq of ⁹⁰Y-1,4,7,10-tetraazacyclododecane-N,N′,N′′,N′′′-tetraacetic acid-d-Phe¹,Tyr³-octreotate (second cycle) shows a consistent accumulation of activity in the suprarenal lesion. Patient 2 suffering from inoperable pancreatic tumour was administered with 2834 MBq of ⁹⁰Y-PRRT (seventh cycle). Patient 3 at his eighth cycle of PRRT (1426 MBq) presents wide areas of radiopharmaceutical uptake within the liver at the level of the third and fourth hepatic segment Reprinted from Fabbri et al⁵² with permission from Mary Ann Liebert, Inc.

⁹⁰Y-PRRT imaging by PET will allow computing tumour and kidney absorbed doses after each therapy cycle. Afterwards in the following cycles, the delivered activity could be tuned in order to optimize therapy. Contrary to Bremsstrahlung SPECT, ⁹⁰Y-PET imaging can be performed in patients receiving a lutetium-177Lu-⁹⁰Y cocktail.

DEVELOPMENT OF NOVEL ACQUISITIONS

Dual radiotracers acquisition

Two different methods are proposed to perform dual radiotracers acquisition in PET: combined use of one “pure” and one “dirty” isotope, or combined use of two tracers of sufficiently different pharmacokinetics.

Andreyev and Celler⁵⁴ proposed to use a “pure” positron emitter (such as ¹⁸F, carbon-11 (¹¹C), ¹⁵O) simultaneously with a “dirty” positron emitter, i.e. that which in addition emits prompt single γ-rays [such as yttrium-86, iodine-124, sodium-22, copper-60 (⁶⁰Cu)]: triple coincidences identify the dirty isotope.⁵⁵

The acquisition of pure positron emitter will be contaminated by dirty decay events when no prompt single γ-ray is detected as a result of system sensitivity or of absorption in the patient. A fraction of the dirty isotope acquisition can be removed from the pure isotope acquisition. This fraction can be computed or measured in phantoms, as performed in dual-isotope SPECT where the highest energy γ-rays are scattered down into the lowest energy acquisition window.

Tumour hypoxia and metabolism could be simultaneously imaged using ⁶⁰Cu-diacetyl-di(N4-methylthiosemicarbazone) co-injected with ¹⁸F-FDG. Another application is the simultaneous study of myocardial hypoxia and perfusion using ⁶⁰Cu-pyruvaldehyde bis(N4-methylthiosemicarbazone) and ¹¹C-acetate. Performing the two acquisitions simultaneously will be more comfortable for the patient, will reduce the patient irradiation by requiring only one radiographic CT, will reduce the cost and will avoid imprecision due to possible metabolic changes between two sequential scans.

With the increasing number of tracers labelled with ¹⁸F, an older method is regaining interest.⁵⁶ The method consists in performing dynamic PET imaging after simultaneous, or staggered, injections of tracers having different pharmacokinetics. The development of sophisticated kinetic models is still in progress in order to separate the contribution of each tracer while keeping a good signal-to-noise ratio in tissues.^57–60

Inline positron emission tomography in hadron therapy

Similar to radionuclide therapy, hadron therapy uses a vector to transport the ionizing energy into the tumour, by the way limiting the irradiation of the surrounding tissues. The main difference is that in hadron therapy, the vector transporting the energy is the hadron itself! Indeed Bragg⁶¹ observed in 1903 that unlike electron and photons, heavy charged particles deliver most of their energy at the end of their range. This range only depends on the particle energy and on the tissues composition. In principle, it is sufficient to choose the right energy and direction of the ions to accurately “paint” the tumour. This is a significant benefit vs internal radiotherapy where the tumour-absorbed dose depends on hardly controllable pharmacokinetics.

Several limitations hamper this ideal scenario. Currently based on CT scan, the therapy planning requires a problematic rescaling of X-ray attenuation coefficient into hadron stopping power. Patient position mismatches between CT and therapy can also introduce some deviations from the planned field. This is especially true in tumours inside soft tissues, such as the neck, which can also move on their own. Techniques to assess in real time the location of the energy deposition would greatly help to tune the Bragg peak at the right place.

In hadron therapy, short half-life positron emitters, i.e. ¹⁵O (122 s), ¹¹C (20 min) and nitrogen-13 (10 min), are produced in the patient's body with a very low abundance. These emitters are commonly imaged with commercial PET systems.^62–64 However, patient transportation from the hadron facility to the PET room spends several tens of minutes. Activity distribution can change due to biological washout during the transportation, making this method impracticable to monitor and to tune the Bragg peak position (Figure 7).⁶⁵

Figure 7.

Repeated scans of the second subject showing the retaining of activity in soft tissue/brain region for short in-room positron emission tomography (PET) scans. The subject was scanned twice, 1 week apart, for two fractionated sessions of the same treatment plan. The list-mode data were reconstructed for the first 5 min or the entire 30 min. The high activity in a soft-tissue region in the 5-min short scan, as shown in the black ovals overlapping with the PET images, decreased dramatically for a 30-min long scan. The corresponding activity profiles (solid for short 5 min, dashed for long 30 min, blue for the first and red for the second scans) and the CT profile (green dotted curve) are shown on the right. For regions with bony structures and fat tissues, the activity was well retained for both short and long scans (first peaks from the right), whereas for regions with brain or soft tissues, the highest peak observed in 5-min scans (second from the right) almost disappeared in 30-min scans. Reprinted from Zhu et al⁶⁵ with permission from the Institute of Physics Publishing.

To overcome these issues, dedicated in-line PET systems are under development and already evaluated in clinical hadron therapy. Early systems were made of dual planar detectors which is an easy method to get an opening for the hadron beam.^66,67 Partial ring detectors were also used.^68–70 In addition to sharing a low sensitivity, these systems required a rotation around the patient to avoid reconstruction artefacts and were thus not optimal for hadron beam tuning. Alternately, separated dual full-ring detectors,⁷¹ slanted⁷² and axially shifted⁷³ single full-ring detectors have been proposed in order to improve the sensitivity while leaving a beam access. TOF information reduces the artefacts when using partial ring geometry:^74–78 two-third ring scanner with 600 ps TOF resolution provides accurate proton dose monitoring. Improved TOF resolutions should allow starting the therapy optimization sooner after the beam lightening is on.

However, positron emitters are not exactly produced at the Bragg peak: significantly before for proton and closer for carbon ions.⁷⁷ Fine therapy optimization, such as hypoxia painting, still requires accurate modelling of these effects.

CONCLUSION

Since its introduction in the seventies, PET has continuously benefited from hardware improvements. At the same time, thanks to the still increasing speed of computers, reconstruction algorithms have also continuously evolved with better corrections for the two major hampering effects, i.e. the scattering of gamma rays in the patient body and the Poisson noise present in the acquired data. The first hampering effect was further reduced by the use of the bright LSO crystal since 2000. Noise artefact in reconstructed image has also been better controlled with the introduction of TOF information made possible by the high LSO speed in 2005.

Current fundamental research focuses on the improvement of TOF resolution, by optimizing the LSO-photosensor block design and by developing new crystals still faster than LSO. Clinically, two emerging novel applications of PET are under development: post ⁹⁰Y radionuclide therapy imaging and dose distribution visualization in hadron therapy. These two applications mainly performed as a post-therapy will likely be more and more involved in therapy optimization. On the other hand, dual tracer PET imaging still requires improvements in order to be clinically usable.