Keywords

14.1 Introduction

The first bremsstrahlung imaging was performed early in 1966 by Simon et al. [1, 2] using a rectilinear scanner following a liver radioembolization via the hepatic artery with 90Y loaded 15-μm-diameter microspheres. Although of very low quality, this imaging already provided the main information: the carcinoid tumours were well preferentially targeted (Fig. 14.1).

Fig. 14.1
figure 1

Tumours appear cold in the diagnostic scan using 198Au-chloride (1 μm-diameter) injected intravenously and trapped in the reticuloendothelial cells of the liver. (Figure reprinted from [2] with permission of the Radiological Society of North America)

Full size image

The first use of an Anger camera for bremsstrahlung imaging was reported by Kaplan et al. in 1985 [3] for 32P. Since the nineties, bremsstrahlung imaging is widely performed in clinical routine for post-therapy check using beta emitter, such as 90Y synovectomy or 90Y liver radioembolization. While medical publications about bremsstrahlung imaging are sparse up to the first decade of the twenty-first century, they significantly increased during the last decade (Table 14.1). This is to be linked to the expanded amount of performed 90Y liver radioembolizations and developments of (faster) Monte-Carlo (MC) numerical methods to improve bremsstrahlung SPECT reconstructions.

Table 14.1 Sorted articles counts obtained from https://pubmed.ncbi.nlm.nih.gov/ when searching the combination (bremsstrahlung AND (SPECT OR SPET OR PET OR planar OR imaging OR whole-body)) in title/abstract
Full size table

Recently, Lhommel, Walrand et al. showed that the low positron emission of 90Y can be usefully imaged by PET after liver radioembolization [4, 5] and after 90Y-DOTATOC therapy as well [6]. Although giving final images a little bit noisier than bremsstrahlung SPECT, pure commercial TOF-PET systems directly provide good spatial resolution and quantification accuracy in 90Y imaging. On the contrary, to achieve similar quality, bremsstrahlung SPECT requires sophisticated reconstruction or acquisition software not yet commercially available [7,8,9]. Those features explained the increasing use of PET to image 90Y in place of bremsstrahlung SPECT at the begin of last decade (Table 14.1).

However, PET modality is not always easily accessible and as 32P and 89Sr do not own any usable isotope emitting γ rays which could be imaged, bremsstrahlung SPECT remains of paramount importance, as shown by the renewed interest in the last years (Table 14.1).

14.2 Bremsstrahlung SPECT Issues

Accurately imaging 90Y, 32P or 89Sr with a γ camera is one of the most challenging topics in nuclear medicine. The bremsstrahlung X-rays are spread along a continuous spectrum extending to energies up to the maximal beta energy emission, i.e. 2.3 MeV, 1.7 MeV and 1.5 MeV for 90Y, 32P and 89Sr, respectively. The maximal energy usable by γ camera owning a mechanical collimator, such as Anger or CZT cameras, is limited to about 0.5 MeV. As a result, all acquisitions using such cameras are inevitably corrupted by high-energy X-rays scattering down into the acquisition energy window.

This high-energy X-rays down scattering contamination includes five different effects (Fig. 14.2): (1) the scattering inside the patient body, (2) the scattering through a collimator septa (usually called penetration), (3) the scattering from a collimator septa, (4) the lead fluorescence Kα and Kβ emissions, (5) the back-scattering from the PMT, electronic boards and lead housing of the camera. Although less frequent, some X-ray paths can include several of the sub-cited effects. Lastly, the β range in the patient also slightly alters the final spatial resolution.

Fig. 14.2
figure 2

The different X-rays paths producing a photoelectric effect in the camera crystal

Full size image

Monte-Carlo simulations allow assessing the individual contributions to the total X-rays producing a photoelectric effect in the camera crystal [10] (Fig. 14.3). Note that contrary to conventional γ emitters, the primary photons represent only a small part of the total detected counts. The most favourable ratio is obtained around 100 keV explaining why the 50 → 150 keV energy window is often chosen [10, 11]. A medium energy general purpose (MEGP) parallel hole collimator is a good choice to reduce the collimator penetration. However, even with this energy window and collimator choice, spatial resolution and quantification accuracy of bremsstrahlung SPECT remain low. Different approaches are proposed to improve this situation: physical effects modelling and adapted collimator design.

Fig. 14.3
figure 3

90Y X-ray energy spectra according to photon origin for a point source 5 cm deep in water cube, MEGP collimation, 3/8 in. thick NaI, computed by Monte-Carlo simulation. Courtesy of Dr. S Heard. The primary to total counts curve (dashed black curve) was added by the author of the present chapter

Full size image

14.3 Intra-Patient Scatter and Collimator-Detector Response Modelling

With the increasing speed of the computer stations, most tomographic reconstructions in nuclear medicine are nowadays performed using the iterative algorithm OSEM, an accelerated version of the EM-ML. Compared to analytical reconstruction, such as FBP, these iterative algorithms have the benefit to correctly account for the Poisson nature of the statistical noise present in radioactive counts measurement and to avoid apparition of negative voxel artefacts. In addition, iterative algorithms can reconstruct any tomographic acquisition setups, i.e. all physical effects introduced in the projection step of the iterative loop will be corrected during the reconstruction process. The state of the art in bremsstrahlung SPECT is thus to model the different X-ray paths during this projection step. However, for the time being an exact modelling of these effects lead to reconstruction time incompatible with the daily SPECT routine.

Minarik et al. [12] built a pre-calculated collimator-detector response (CDR) table by Monte-Carlo simulation and modelled the scatter into the object using the effective source scatter estimation (ESSE) method [13]. The modelling was performed for a 105–195 keV acquisition window in order to avoid the lead fluorescence X-rays contamination. This model was incorporated into the OSEM reconstruction algorithm. Evaluation in an abdominal phantom showed a quantification accuracy of 8.5% for the liver activity. Accuracy in lesions activity measurement was not assessed.

This group applied their correction method in three patients receiving high-dose 90Y radioimmunotherapy [14]. The patients were imaged at 1, 24, 48, 72, 144, 166 h by SPECT/CT after a pre-therapeutic injection of 300 MBq of 111In-ibritumomab. Patients received a 90Y-ibritumomab activity computed to deliver 12 Gy to the liver based on the pharmacokinetics measured with the 111In SPECT/CT and were imaged at the same time points post injection by bremsstrahlung SPECT/CT. The absolute relative differences between organ absorbed dose computed from 111In and 90Y SPECT/CT were 8.8 ± 13.7, 8.9 ± 4.0 and 51.7 ± 18.9 (mean ± std in %), for the liver, kidneys and lungs, respectively. This showed that this bremsstrahlung SPECT/CT correction method can be used for the body region below the lungs: for the time being, the ESSE method does not account for tissues density variation, such as present in the slices crossing the lungs.

Elschot et al. [15] developed a quantitative Monte-Carlo based SPECT reconstruction for 90Y applications. They implemented a Monte-Carlo simulator to model the photon attenuation and scatter for the full 90Y spectrum in the projection step, while pre-calculated convolution kernels were used for the collimator-detector response. Using a 50–250 keV energy window, they obtained quantification improvements compared to standard clinical SPECT-CT reconstructions for NEMA image quality phantom, and results close to PET-CT for a small set of patients after radioembolization with 90Y resin spheres.

In order to improve the count rate for applications where the activity is modest, such as 90Y radioimmunotherapy, Rong et al. [16] computed by Monte-Carlo the CDR and ESSE for the wide 100–500 keV acquisition window. The computation was performed using separate treatment of photons in various energy ranges and in various logical categories. Evaluation in an elliptical phantom containing three spheres of 5.5 cm, 3.3 cm and 1.5 cm-diameter, with specific activity ten-, ten- and 20-fold that of the surrounding background, showed a quantification accuracy of 7%, 9.7% and 10.2%, respectively.

To better take into account the energy dependence of photon–matter interaction probabilities and to simplify scatter corrections, Siman et al. [17] adapted the standard multi energy window approach to correct 90Y SPECT-CT images for scatter. After splitting the 70–410 keV energy range into six windows whose widths were selected to enable a single attenuation coefficient use per window, they determined from phantoms the 90–125 keV and 310–410 keV ranges to be the best windows for imaging and scatter respectively. The scaling factor to multiply the scatter image was estimated from planar acquisitions of patients. This approach, while improving image quality and quantification, is strongly dependent on the camera and collimator used, and approximate as the scatter factor was averaged over different patients.

Dewaraja et al. [18] made use of Monte-Carlo only to estimate the scatter of photons, that they combined with a 3D OSEM reconstruction with an analytic projector for 90Y SPECT-CT. Using a single energy window of 105–195 keV and a high-energy collimator, they showed that two iterations for the scatter estimation were enough, allowing a reduction of the reconstruction time to about 40 min. They so better recovered the activity in a liver/lung phantom, with values of 86%, 104% and 104% for the recovery in the intrahepatic lesions, normal liver and lungs respectively. They similarly observed an increase of the lesion to liver concentration with their MC reconstruction for patient studies compared to standard SPECT-CT reconstructions.

More recently Chun et al. [19] proposed an interesting approach of joint spectral reconstruction for quantitative 90Y SPECT imaging. They used multiple narrow acquisitions windows with multi-band forward modelling. The latter has the advantage to permit a better modelling of energy dependent physics like using more adequate attenuation coefficients in place of only one corresponding to the mid-range energy. The former allowed the authors to extract and combine activity data from the different energy window acquisitions to improve the reconstructed image. They also presented an accelerated algorithm using energy subsets similarly to the angular subsets in OSEM. MC simulations were used at every five iterations to estimate the scatter components without impacting too much the reconstruction time. With this new algorithm they reached faster convergence of the recovery coefficients in phantom studies than with single spectral/single energy window and multi spectral/single energy window reconstructions. The proposed method has also the advantage to implement matched forward and backward projectors, inducing better convergence stability, unlike most of developed reconstruction algorithms where MC scatter and collimator modelling is only incorporated in the forward projector.

Another physical characteristic seldom considered is the dependence of the bremsstrahlung generation with respect to the tissue types. Lim et al. [20] modelled by Monte-Carlo the bremsstrahlung generation probabilities according to the tissue. With their adapted reconstruction algorithm, they showed on phantom acquisition and simulations improved estimations of 90Y activities in tissues differing from water. Unfortunately that study did not include absolute quantification, nor patient imaging.

Very recently, Xiang et al. [21] made use of latest developments in artificial intelligence to train a deep convolutional neural network from 90Y SPECT projections and CT attenuation maps to produce scatter projections. The training set was made of simulated numerical phantom data for which actual scatter component is known. With that approach they obtained image quality and contrast recovery similar to MC scatter estimation but in much less time. This still requires developments in training data and quantitative validations, but could pave the way to better bremsstrahlung SPECT-CT reconstructed images in clinical routine.

However, in medical imaging it is always profitable to improve the hardware performance in order to acquire the right events, rather than to correct for the contaminating events afterwards. Indeed, this last solution inevitably results in a higher noise level regarding the statistical nature of primary and contaminating photon. Amazingly, other choices of collimators than MEGP have been considered only very recently.

14.4 Choice of Collimators Better Adapted to Bremsstrahlung Imaging

Van Holen et al. [8] proposed the use of a rotating slat collimator that owns a much higher geometric efficiency than a parallel hole collimator. As a result the relative importance of septal penetration is reduced, resulting in better contrast to noise ratio. Note, that regarding only the primary X-rays, the high geometric efficiency improvement of the rotating slat collimator is counterbalanced by less information provided about the X-ray coming direction. Data publication in a full paper is still pending.

Walrand et al. [9] used a conventional thyroid dedicated medium energy pinhole (MEPH) collimator in bremsstrahlung SPECT in purpose of 90Y liver radioembolization check. Scattering inside the phantom was modelled using an adaptation of an effective scatter model previously developed for 99mTc [22], similar and anterior to ESSE [13]. Compared to parallel hole collimators, the high-energy X-rays can hit lead material mostly only on the external side of the pinhole collimator housing (Fig. 14.4a). As a result, the lead fluorescence and scattering X-rays cannot reach the crystal as the collimator housing thickness (2 cm) is sufficient to significantly reduce penetration. Disturbing penetration-scattering can only occur on the small tungsten insert, the fluorescence X-ray emissions of which are located below 10 keV [23]. These features improve the primary to total count ratio (Fig. 14.3). The comparison between 90Y and 99mTc radial profile in air (Fig. 14.4b) shows that using a 50–150 keV acquisition window the primary photons represent 68% and 31% of the total detected photons for MEPH and MEGP collimators, respectively.

Fig. 14.4
figure 4

(a) reduction of the disturbing X-rays paths contributing to the total crystal photoelectric counts. (b) point spread function (PSF) comparison of a 90Y and 99mTc point source in air with a MEPH and MEGP collimator using a 1/2 in. thick NaI GE 400 AC camera with a 50–150 keV acquisition window. 99mTc PSF approximates the PSF of the primary photons in the 90Y acquisition

Full size image

Evaluation in cold and hot spheres phantom [9] showed that MEPH SPECT provided quantification accuracy similar to that of TOF-PET, but with significantly less noise. Helical MEPH SPECTs of a realistic liver radioembolization phantom were also acquired and showed that reproducible accurate activity quantification can be obtained in 1 min acquisition time (relative deviation of healthy liver compartment: 10 ± 0.1%).

Gupta et al. [24] showed the feasibility of real-time visualization of iron-labelled microspheres delivery during liver SIRT in rabbits using MRI. In this paper, cosigned by R. Salem, the authors concluded: “Although quantitative in vivo estimation of microsphere biodistribution may prove technically challenging, the clinical effect could be enormous, thus permitting dose optimization to maximize tumour kill while limiting toxic effects on normal liver tissues.” However, human liver SIRT appears quite incompatible with MR: the X-ray angiographic imager will difficultly be implemented around the MR table, and the long duration of liver SIRT, that can overpass 2 h for challenging arterial trees, is not supportable by most of MRI agenda.

14.5 Current Bremsstrahlung SPECT-CT Routine Applications

For the time being, dedicated bremsstrahlung acquisition hardware or reconstruction software are not commercially available, and most bremsstrahlung routine SPECT-CT are performed without any special correction. This is not a major problem as far as they are intended to post-therapy visual check. However, users have to be very cautious when performing quantitative measurement on standard acquisition-reconstruction of bremsstrahlung X-rays.

90Y radio-synovectomy has been the main application of bremsstrahlung routine imaging since 30 years. However, regarding to the small size of the synovial compartment, the higher spatial resolution of PET system is a major benefit to perform an optimal check of this therapy [25]. In addition, as the activity is located into a small volume, noise issues are very limited.

Liver radioembolization with 90Y loaded microspheres by catheterization through the hepatic artery is an emerging treatment for primary and metastatic liver cancer. However selective radioembolization fully confined in the targeted hepatic artery branch is a challenging operation. Microspheres can be spread in the arterial tree along a different pattern than those of the macro-aggregates or of the contrast agent (see “choice of a surrogate” in chapter “SPECT for dosimetry”) leading to adverse effects [26].

Until the possibility of tracking the microspheres delivery during the catheterization, it is thus of paramount importance to check the microsphere distribution post therapy in order to take the appropriate cares to reduce the side effects in case of activity delivered in critical tissues. Ahmadzadehfar et al. [27] evaluated the significance of bremsstrahlung SPECT/CT in the prediction of extrahepatic side effects after 188 radioembolizations with 90Y-microspheres. They observed a dramatic improvement of the sensitivity and of the positive predictive value of SPECT/CT (87% and 100%) compared to SPECT alone (13% and 8%), leading to a final accuracy of 99%. The two cases shown in Fig. 14.5 clearly illustrate the benefit of this co-acquisition.

Fig. 14.5
figure 5

90Y bremmstrahlung SPECT (a)/CT (c) co-registration (b). Top row: Patient with a focal activity in the duodenum as shown by the SPECT-CT co-registration (thin yellow arrow). Patient received a daily pump inhibitor, but got a duodenal ulcer without active bleeding confirmed by gastroduodenoscopy. Bottom row: Patient with a suspicious focal activity in SPECT (thick yellow arrow), but proved to be hypertrophied left liver lobe in the SPECT-CT co-registration. (Reprinted from [27] with permission of Springer-Verlag)

Full size image

Peptides receptor targeted therapies are usually performed in several cycles in order to limit kidneys toxicity [28]. Fabbri et al. [29, 30] evaluated on anatomical phantom the feasibility to perform organs dosimetry after each 90Y-DOTATOC cycle in order to optimize the activity to be injected in the next one. They showed on a phantom that, using the standard reconstruction software of the SPECT/CT system, but with calibration factors depending on the lesion or organ volume measured on the CT, it was possible to access the dosimetry within an accuracy of 10%. Calibration factors ranged from 0.4 to 1.3 for the tissue ranging from 8 to 150 mL. However this method does not correct for the cross scattering organ contamination which can be problematic for the activity quantification in kidneys with close bowel or tumour surrounding activities. Studies on phantom of various sizes and of various activity distributions are thus suitable to further assess its accuracy.

Following a clinical study with 90Y Zevalin to treat lymphomas, Shiba et al. [31] analysed the impact on 90Y bremsstrahlung SPECT images of the presence of residual 111In from pre-therapeutic evaluation. Their results indicated that 111In is still prevalent in the post-therapeutic image 1 week after its injection. Combined with the limited image quality of 90Y bremsstrahlung SPECT, the authors recommend the use of PET/CT imaging for that kind of application.

14.6 Dose–Response Studies Based on Bremsstrahlung SPECT

With the speed-up development in the last decade of computer tools that improve the image quality of bremsstrahlung SPECT/CT, several groups started performing dosimetric studies based on bremsstrahlung and were able to extract dose–response relationships.

Kappadath et al. [32] performed voxel dosimetry based on bremsstrahlung SPECT/CT on patients treated for hepatocellular carcinoma by 90Y-radioembolization with glass microspheres. From SPECT/CT iterative reconstructions including attenuation, scatter and collimator modelling, they obtained a correlation between dose metrics and mRECIST response, with a mean dose of 160 Gy and a mean biological effective dose of 214 Gy for 50% probability of response, giving positive predictive value of 70% and negative predictive value of 62%.

Piasecki et al. [33] studied the dose–response for colorectal liver metastases after 90Y radioembolization. They obtained relationships between predicted tumour doses from 99mTc-MAA SPECT/CT and responses, but not when looking at the actual 90Y tumour absorbed doses evaluated from 90Y bremsstrahlung SPECT/CT. This may be in part due to the use of standard iterative reconstruction that does not include dedicated scatter or collimator modelling. Using standard manufacturer reconstruction without specific corrections. Schobert et al. [34] were able to obtain dose–response relationship in HCC after 90Y radioembolization, but not for non-HCC lesions, by performing bremsstrahlung SPECT/CT with low energy high resolution collimators and standard iterative reconstruction.

These studies should be looked at as early and promising examples of feasibility of dosimetric analyses, but many developments and validations are still needed before accurate dosimetry can be obtained from bremsstrahlung SPECT/CT imaging.

14.7 Perspectives: New Detectors Better Adapted to Bremsstrahlung Imaging

Cadmium zinc tellurium (CZT) detectors, although still expensive, are emerging in dedicated cardiac SPECT systems [35] where their compactness and absence of dead edge area allows to build area of independent detectors all focussing to the heart. With the future decrease in manufacturing cost, CZT will become more and more used for general purpose γ cameras in order to profit of its better energy resolution [36]. Indeed, for γ emitters, a better energy resolution allows to narrow the acquisition window around the photoelectric peak which reduces the detection of γ rays scattered inside the patient body and results in a better image contrast. These CZT cameras should also be better adapted to bremsstrahlung imaging as the quantity of medium Z material (glass, iron, copper), present between the γ detection area and the detector housing, and which backscatter the high-energy X-rays, will be reduced. The camera housing should preferably be made in tungsten which has its fluorescence X-rays emission around 10 keV [23].

Recently Walrand et al. [37] simulated the detected energy spectra for 90Y to illustrate the impact of scattered bremsstrahlung rays. Figure 14.6a represents the signal to scatter ratio, i.e. the total counts coming from the geometric X-rays to those coming from scattered X-rays, with respect to the detected X-rays energy. The signal clearly increases when the NaI crystal is replaced by a BGO crystal, and again when the high-energy parallel hole collimator is replaced by a pinhole collimator.

Fig. 14.6
figure 6

(a) Monte-Carlo simulations of detected energy spectra for a 90Y source in air in front of a gamma camera including a NaI crystal and equipped with a high-energy general purpose (HEGP) collimator (red curve), including a BGO crystal and equipped with a HEGP (green curve) or a high-energy pinhole (HEPH) collimator (blue curve) [37]. The curves represent the ratio of geometric counts to scattered counts with respect to the detected ray energy. (b) Illustration of a mobile dual head BGO-camera equipped with HEPH collimators to be used directly in the catheterization room to image the microspheres distribution and help the radiologist in optimizing the patient dose

Full size image

The simulation of the BGO-pinhole camera shows that the ratio is above 1 (geometric X-rays outnumber scattered ones) in almost all the 50–500 keV energy range. This opens the way to the development of the continuous energy tomography where the detected X-rays energy provides additional information for the image reconstruction similarly to the detector angle in conventional SPECT imaging.

This could be used to optimize the patient dose directly in the catheterization room (Fig. 14.6b). After a first injection of microspheres (90Y, 166Ho or 32P), a BGO-pinhole camera is moved around the patient and a 2 min scan provides data that are treated by a dosimetric software. From its results the radiologist can estimate if and how much activity can still be injected in the patient. The procedure can be iterated until optimal activity has been injected.

Compton cameras are under intensive development in physics labs [27]. Similarly to PET, the purpose of Compton camera is to get free of the mechanical collimator in order to improve both the spatial resolution and the sensitivity as well. Applied to bremsstrahlung imaging, Compton camera will be free of any collimator disturbing effects. In Compton camera the photon is detected in two distant planes, i.e. the Compton and the scintillation detector. Thus, measuring the time detection sequence between these two planes will also allow discrimination between primary and backscattered photons. Lastly, sensitivity of Compton camera increases with rays energy, and acquisition with very wide energy window could be used to further increase the sensitivity.

14.8 Conclusions

With the increasing accessibility to PET systems, 90Y PET imaging will likely remain the gold standard for 90Y imaging. However the increasing amount of performed 90Y therapies will see the continuing developments and improvements of 90Y bremsstrahlung SPECT imaging, especially on the software side with better corrections to achieve imaging quality similar to PET. Moreover, bremsstrahlung SPECT will remain the only way to image 32P and 89Sr, two isotopes regaining interest in radiotherapy. In addition, use of a dedicated pinhole collimator could allow performing fast bremsstrahlung SPECT quantification during liver radioembolization. Soon, the emerging CZT detectors should further improve bremsstrahlung SPECT. In a distant future, Compton camera could revolutionize this imaging modality.