Absorbed dose calculations for tumours and for normal organs-at-risk follow essentially the same protocol. Multiple whole-body scans have frequently been used for patient dosimetry, and these have the advantage that the full distribution of activity within the body can be visualised. However, to perform the most accurate dosimetry possible for 131I mIBG therapy, SPECT or, ideally, SPECT/CT imaging is highly recommended due to the variable localisation of tumours and organs which can entail substantial contributions from active uptake in under and overlying tissues. This is particularly true for larger lesions which can also demonstrate heterogeneous uptake. As gamma cameras are not optimised for high-activity imaging of high-energy gamma emitters, several procedures must be performed to determine the activity imaged in each scan prior to performing dosimetry calculations.
Quantification/calibration
Image quantification, particularly for 131I, has been the subject of significant research for many years. Nevertheless, it has been demonstrated that 131I imaging data may be quantified with sufficient accuracy to produce clinically meaningful results [5, 25]. A number of variations have been employed to calibrate the count data acquired [25, 27,28,29,30,31]. One approach is to obtain sensitivity measurements from a point source in air and use this sensitivity factor in the reconstruction to obtain an image in units of activity concentration. However, the sensitivity measurement is sensitive to septal pentetration which varies with source to collimator distance. The reconstruction accuracy is also highly dependent on corrections for scatter, attenuation, system spatial resolution and collimator detector response. For this reason, it is recommended to calibrate post-reconstruction with a volumetric phantom (such as a 20-cm-diameter Jaszczak cylinder) filled with a known concentration of activity [32]. The advantage of this approach is that it accounts for variations in local protocols for image acquisition and reconstruction software. For the calibration protocol, it is essential that the 131I activity, concentration and phantom volume are accurately known and, therefore, stock volumes should be measured in an activity meter calibrated to national standards. Activity within the phantom should be sufficiently low to ensure that there is no relevant dead-time-related count losses during scanning. It is essential to image and process the calibration phantom data in the same way as the patient data (employing the same reconstruction parameters including scatter and attenuation correction) [31].
The camera calibration factor, Q, is determined from the count rate within a volume of interest (VOI), 75% the physical phantom size, centred within the reconstructed phantom image volume, divided by the exact activity concentration at the time of scanning and the VOI volume;
$$ Q=\frac{{\dot{C}}_{\mathrm{VOI}}}{A_{\mathrm{VOI}}}, $$
(4)
where \( {\dot{C}}_{\mathrm{VOI}} \) is the count rate measured in the VOI, and AVOI is the decay-corrected activity in the VOI, i.e. the activity concentration multiplied by VOI volume. The size and position of the VOI should be chosen such that the observed count rate is not affected by partial volume effects or statistical noise. This can be tested by using multiple small VOIs within the larger VOI volume and ensuring that a similar value for Q is obtained in all cases.
Partial volume correction
In addition to the sensitivity measurement, it is also recommended to characterise partial volume effects by imaging a series of phantoms with different volumes and levels of activity that mimic the range for which dosimetry may be carried out. This phantom series should comprise a set of fillable spheres or cylinders of known volume (ideally up to 60 mm in diameter for 131I-based imaging which is typically performed with high energy collimation), placed within the larger phantom. Care should be taken to ensure minimal cross talk of counts between inserts, which may require larger inserts to be scanned separately. A recovery term for each insert volume, which can be used to volume-dependently correct for partial volume-based count losses, is given by
$$ {R}_i=\frac{{\dot{C}}_i}{QA_i} $$
(5)
where \( {\dot{C}}_i \) is the observed count rate measured within a VOI matching the true volume of the insert i and Ai is the known activity within insert i. When subsequently applying the recovery factor to clinical data, any difference between the shapes of the VOI to that of the recovery phantom will influence the accuracy of this correction, as will differing target-to-background concentrations. In some instances, it may be more appropriate to use anthropomorphic-shaped inserts specific to the patient.
Dead time
The high count rate encountered from patients treated with 131I can cause camera dead-time-related count losses. This is particularly pertinent when imaging during the first days post administration, whereby the counts acquired by the camera do not increase linearly with increasing activity [31, 33, 34]. The behaviour of cameras can vary widely so it is essential to characterise the camera used for therapy imaging. A dual source method has been proposed [35], whereby the count rate from a known low activity source is measured with and without the presence of a high activity source. This approach is quick; however, it does not account for the contribution of scatter events that can affect system dead time. An alternative approach entails placing a known source within the field of view and normalising the total counts acquired within the image to those obtained from the source without the patient present. However, this is unwieldy and subject to errors from interference of counts emanating from the source and the patient. It is therefore recommended to characterise the camera for dead time by imaging a source in a scatter medium (such as a water-filled Jaszczak phantom) at varying activity levels [32]. This can be achieved from static acquisitions of a decaying source, although this requires many measurements over a period of weeks. A more practical solution is to repeatedly image a source to which small quantities of known activity are continually (and carefully) added. A detailed description of this methodology and comparative characteristics of different manufacturers is given by [31]. For a paralysable system, the observed count rate measured by a gamma camera, \( {\dot{C}}_{\mathrm{obs}} \), is related to the incident count rate, \( {\dot{C}}_{\mathrm{inc}} \), and the system dead time, τ, as
$$ {\dot{C}}_{\mathrm{obs}}={\dot{C}}_{\mathrm{inc}}{e}^{-{\dot{C}}_{\mathrm{inc}}\tau } $$
(6)
The methodology employed to measure dead time from a decaying source is similar to that described by the National Electrical Manufacturers Association (NEMA) [36] to determine count rate performance. Dead time characterisation obtained from increasing source activities are determined by linearly extrapolating the measured count rates at low activities to that at higher activities. A nonlinear least squares minimisation algorithm can then be used to determine an estimate of τ for all values of \( {\dot{C}}_{\mathrm{obs}} \) and \( {\dot{C}}_{\mathrm{inc}} \). In addition to the measurement of τ, it is necessary to determine potential changes in image uniformity that can arise from high count rates. An illustrative example of this is given in Fig. 1 for a uniform cylindrical phantom imaged at low and high count rates. In this example, although an absolute correction for count losses can be made, the tube artefact effect on the high count image renders the image unsuitable for quantitative imaging.
In patient acquisitions, correction for dead-time-related count losses can be applied to each projection, based on the count rate incident at each projection angle. Alternatively, an average correction can be applied (pre or post reconstruction) based on the mean count rate incident over all projections. Correction of the dead-time-related count losses in each pixel, voxel or VOI count rate should then be multiplied by a dead time correction factor, DTF, which can be determined iteratively [37] using the measured system dead time and the observed count rate in the patient projection data, \( {\dot{C}}_p \).
$$ {\mathrm{DTF}}^{i+1}={e}^{\left({\mathrm{DTF}}^i\times \tau \times {\dot{C}}_p\right)} $$
(7)
Data acquisition
Although the design and performance of gamma cameras has changed little in the last 25 years, there are nevertheless differences between systems that can affect results. For example, thicker crystal thicknesses are available for some systems (typically 15.9 mm rather than 9.5 mm) which may render the system more suitable to pre-therapy tracer studies and later-time point therapy imaging but oversensitive to early time point high activities. This may impact the quantitative accuracy of scans acquired shortly after a therapeutic administration. In general, it is recommended that patients are imaged on the same scanner for the entirety of the study. Due to septal penetration of the high-energy photons (364 keV, 637 keV and 723 keV) emitted by 131I, a high-energy general purpose parallel hole or ultra-high-energy collimator should be used [38]. Energy windows should be configured to enable triple energy window (TEW) scatter correction to be performed, consisting of a 15–20% window around the main energy peak and 6% windows placed on either side and immediately adjacent to the main window. Wider scatter window widths are not recommended as they may increasingly include scatter events spatially inconsistent with that inside the photopeak, and for the case of the lower window partially cover the 284 keV gamma emission. The camera radial position should be set with auto contouring to ensure optimal spatial resolution.
Data should be acquired into a 128 × 128 matrix or higher with a minimum of 60 projections (or 6° angular steps). The acquisition time for each projection will be dependent on the count rate and on patient comfort. Typical scan times may be as little as 5 s per projection for an initial scan or up to 60 s or more for later scans. The timing of the initial scan will be dependent on the camera dead time characteristics, but should be acquired as soon as possible after administration. Ideally, further scans should be acquired at least daily until either the count rate is too low or until the patient has been discharged and is unable to attend the department for further scans. In practice, a minimum of 3 scans should be acquired to ensure a basic minimum accuracy for subsequent fitting to the time–activity curve. A useful measure to test that the TAC is adequately charecterised is to calculate the time-integrated activity over the time period between the first and last measurements. It is recommended that the fraction of the time-integrated activity between measurements is greater than 80% of the total time-integrated activity when extrapolated from zero to infinity [20].
Scatter correction
The raw image data should be corrected for scatter using the data acquired in the three energy windows according to the method of Ogawa et al. [39] whereby the scatter-corrected counts (CSC) in the main window are given by
$$ {C}_{\mathrm{SC}}={C}_{\mathrm{peak}}-\frac{W_{\mathrm{peak}}}{2}\left(\frac{C_1}{W_1}+\frac{C_2}{W_2}\right), $$
(8)
where Cpeak is the projection counts acquired within in the main window of width Wpeak and C1/C2 are the projection counts acquired within the lower/upper energy scatter window of width W1/W2, respectively.
In modern reconstruction software, scatter correction is typically automatic and included in the iterative loop. Older software may, however, require subtraction of projections. In addition, some newer software use a simulated scatter image, negating the need to acquire separate scatter windows.
Attenuation correction
As many neuroendocrine tumours are deep-seated, attenuation correction should be performed according to local protocols. As hybrid SPECT/CT scanners become more widespread, the CT scan can be used for generating an attenuation map, although the lack of CT data should not inhibit the application of dosimetry. If no CT-based attenuation is possible, a relatively simple yet effective method of attenuation correction may be applied according to the method of Chang et al. [40]. For paediatric patients, the abdomen can be considered similar to an adult head and, therefore, the use of a linear attenuation assumption can be justified when compared to common practice for SPECT brain imaging. The attenuation coefficient of 364 keV photons in water is 0.11 cm-1 [32], and a reasonable estimate of the body outline can be visibly obtained from the lower energy window scatter image.
Image reconstruction
Iterative reconstruction methods are recommended. However, it should be noted that results can vary highly depending on the reconstruction parameters used, which should always be clearly reported [41]. Optimal reconstruction parameters for diagnostic imaging do not necessarily provide the most accurate quantification. For this reason, an optimisation of the reconstruction is required. Estimates of the optimal reconstruction parameters can be obtained by examining the calibration acquisition data. Starting from the default clinical settings, the number of OSEM updates (product of iterations and subsets) should be increased and the recovery coefficient, R, series should be determined as a function of updates. Ideal reconstruction settings are those when R is close to the maximum achievable without significant image noise. The noise level can be studied with a VOI in the large cylindrical phantom, plotted as a function of update number. It may be necessary to adjust the number of updates once clinical data are acquired to ensure image convergence and appropriate noise levels. Many modern reconstruction algorithms offer collimator-detector response corrections which can lead to higher count recovery during reconstruction and suppressed image noise. However, care should be taken to avoid Gibbs ringing artefacts that can manifest with this correction and ultimately misrepresent the distribution of activity [42].
Volume definition
Absorbed doses may be calculated for either the entire volume of the tumour or normal organ or for any sub-organ volume. Conventionally, mean absorbed doses are calculated over the entire volume of interest. For this approach, results are highly dependent on delineation of the volume. For visibly uniform uptake and where ‘anatomical’ imaging is available (i.e. CT or MRI) with sufficient contrast to delineate the organ or lesion, this may be used. In cases with heterogeneous uptake or insufficient anatomical contrast, ‘functional’ (based on SPECT/PET data) volume delineation may be necessary. A widely used approach is to use an adaptive threshold approach on the SPECT data using factors determined from the calibration data [43, 44].
Once the appropriate VOI is defined, the count rate at each time point \( {\dot{C}}_v(t) \) is converted to activity using the appropriate calibration, recovery and (if being applied, post reconstruction) dead time factor
$$ {A}_v(t)=\frac{{\dot{C}}_v(t)}{Q\bullet R(v)}.\mathrm{DTF} $$
(9)
Voxel-based dosimetry has been proposed as a method for which calculations may be subsequently determined for any defined volume. Many neuroendocrine tumours are of a substantial volume and contain a heterogeneous distribution of activity, visible even within the limitations of 131I imaging. Sub-volumes of a region of interest may therefore also be defined. Of potential interest are the maximum and minimum absorbed doses. A relatively simple approach (but also prone to noise-related artefacts) is to identify a single voxel in the sequential scan set that represents the maximum or minimum uptake. An extension of this approach is to identify a sub-volume, consisting for example of a block of 27 voxels that represent volumes of maximum or minimum uptake. This larger volume is potentially less prone to noise-related artefacts compared to reporting a single voxel absorbed dose.
Time–activity curve (TAC)
Effective half-lives can vary widely from patient to patient and should not be assumed based on a population average. A series of sequential scans is therefore essential. As SPECT has a limited field of view, it may be necessary to obtain two sets of scans if dosimetry is required for widely separated volumes. The time–activity data for the chosen region should be plotted and the area under the curve should be integrated. The method of integration and the assumptions made regarding uptake or decay before the first time point or following the last time point can have a significant impact on the time-integrated activity calculations and should be addressed carefully. It is recommended that extrapolation of the later time points is performed to define the effective decay phase, rather than assuming physical decay.
Absorbed dose calculation
MIRD S values [45] or RADAR dose factors [26, 46], used to convert the time-integrated activity to the absorbed dose, are available for standardised organ geometries but are not defined for tumours, nor for sub-volumes within tumours or normal organs. For the calculation of absorbed doses to normal organs, such as the liver, the adjustment of mass according to patient-specific measurements can improve the accuracy significantly and is recommended [47]. For non-standard organs (such as lesions) the S value chosen for the absorbed dose calculation will be dependent on the region of interest chosen. In many cases, a spherical S value will be sufficient to provide a reasonable estimate of the absorbed dose. S values for individual voxels (dose point kernels) of varying size are available from a number of sources [48] for several radionuclides. It is possible to perform dose calculation beyond the mean absorbed dose, that take into account dose rate and heterogeneity in tracer uptake, such as the biological effective dose (BED) and equivalent uniform dose (EUD). Methods for calculating these parameters is beyond the scope of this guidance document but is well described within the literature [49].