Patient-specific image-based bone marrow dosimetry in Lu-177-[DOTA⁰,Tyr³]-Octreotate and Lu-177-DKFZ-PSMA-617 therapy: investigation of a new hybrid image approach

This study is based on ten patients, with five patients suffering from somatostatin receptor-positive neuroendocrine metastases (P1-P5) and five patients from mCRPC with expression of PSMA-avid lesions (P6-P10). Details for each patient are provided in Tables 1 and 2. All patients received multiple therapy cycles of approximately 3.7 GBq Lu-177-DKFZ-PSMA-617 (Lu-177-PSMA-617) or 7.4 GBq Lu-177-[DOTA⁰,Tyr³]-Octreotate (Lu-177-Octreotate). All patients except one mCRPC patient showed soft tissue lesions on the pre-therapeutic Ga-68-HBED-CC-PSMA or Ga-68-[DOTA⁰,Tyr³]-Octreotate PET/CT scans, while all prostate-specific membrane antigen (PSMA) patients and two NET patients additionally presented with bone metastases (Tables 1 and 2). The local ethics committee approved the study protocol and did not desire any written consent for the study entry. The study is based on retrospective and anonymised patient data.

Data for dosimetry were acquired during a routine 4-day in-patient stay following the radiopharmaceutical injection, in conjunction with standard clinical examinations. All patients received a 15-min one-bed abdominal SPECT scan and a 20-min whole-body planar scintigraphy at 24, 48, and 72 h post-injection (p. i.) on a dual-headed Symbia T2 SPECT/CT (Siemens Medical Solutions, Erlangen, Germany). Counts were detected for the photopeak window of 208 keV (width 15%) by the usage of a medium-energy low-penetration collimator. Two additional scatter windows were measured at 170 keV (width 15%) and 240 keV (width 10%). A low-dose AC-CT was acquired at the first image acquisition session for anatomical correlation and attenuation correction during quantitative SPECT reconstruction. For the determination of the absorbed dose to the bone marrow from the activity circulating in the blood, five venous blood samples were drawn from the site contralateral to injection at 30 and 80 min p. i. and 24, 48, and 72 h p. i. [9, 20].

Quantitative SPECT images were reconstructed as described by Delker et al. [9] via a rotation-based, penalised, one-step-late ordered subset expectation maximisation algorithm, which included corrections for scatter, attenuation, and distance-dependent geometrical collimator blur. Attenuation correction was performed for each SPECT scan via the AC-CT, which was acquired along with the SPECT scan 24 h post-injection. To apply the attenuation correction, especially to the SPECT scans 48 and 72 h p. i., the single AC-CT was co-registered onto an initial SPECT reconstruction without attenuation correction by using a rigid body co-registration algorithm with six degrees of freedom (PMOD Version 3.609, PMOD Technologies, Zurich, Switzerland). If only one AC-CT is acquired for sequential SPECT imaging, special care has to be taken to minimise misregistration between SPECT images and separately acquired CT scans, as such a misalignment can distort the proper attenuation correction and, thus, activity quantification. This is in principal also true for serial SPECT and CT imaging, as even within a single image acquisition session patient movements cannot be entirely avoided. Scatter correction employed the triple energy window (TEW) method. Correction for distance-dependent collimator blur made use of a Gaussian blur model. Corrections for partial volume effects and dead time were not applied. For conversion of the measured counts per second and per voxel to Becquerel per millilitre, an appropriate calibration factor was determined. Therefore, we used a large cylinder of approximately 20 cm diameter, which was filled with a known activity concentration and which has been imaged and reconstructed via the same protocol [9, 15, 20].

For each patient, all acquired whole-body planar images were corrected for scatter and attenuation on a pixel basis via a dedicated MATLAB routine (Fig. 1) [16, 20, 21]; for the correction of scatter, the TEW method was applied, as for the quantitative SPECT images. For the attenuation correction, a linear projection of μ values along the ventral axis of the patient was created from the diagnostic CT image of the pre-therapeutic Ga-68 PET/CT scan, which covered nearly the whole patient body from the middle of the head to approximately the knees. Therefore, a conversion between the Hounsfield units (HUs) in the diagnostic CT and the μ values at 208 keV was established by acquiring a CT scan of a Gammex tissue phantom (Gammex 467; Gammex, Inc., Middleton, WI) with 16 tissue rods of known composition and thus known attenuation characteristics [22]. The μ values of all rods were plotted against the measured HUs, and a bilinear fit model (range 1: HU = (− 688;0); range 2: HU = (0;1127)) was applied to the whole data set [15, 22]. This calibration curve allows for the assignment of μ values to a continuous range of HUs. The lower arms and legs as well as a part of the head were not included in the PET/CT data, as the arms are usually positioned above the head during the PET/CT scan and as the PET/CT scan is usually not acquired over the entire patient length. By contrast, the arms, legs, and head are fully included in the whole-body planar images, and an appropriate μ value has to be defined for each segment (Fig. 1). Thus, mean μ values derived from three patients with PET/CT acquisitions of the head, legs, and arms were assigned to the missing segments. Therefore, all segments—the part visible on the PET/CT and the missing parts of the head, arms, and legs—were delineated on the co-registered whole-body planar images (delineation and rigid body co-registration via PMOD Version 3.609). The resulting map of the regions of interest (ROI) was saved, with each ROI segment being characterised via a defined value. This ROI map was then loaded by a self-designed MATLAB routine, which assigned the defined μ value to each segment according to the ROI number. Afterwards, the resulting whole-body integral μ-map was blurred via a Gaussian filter with a width approximating an average resolution of the gamma camera (geometric resolution of full width at half maximum of 11 mm at 10 cm). Pixel-wise attenuation correction was finally performed in conjunction with geometric averaging of both planar views (conjugate view method) [21].

Calibration of whole-body planar images was performed via the corresponding quantitative abdominal SPECT by using the fact that ideally the total activity A_SPECT within the quantitative SPECT should be correlated to the number of counts per second (cps) x_planar in the planar abdominal counterpart multiplied by an appropriate calibration factor:

C_{planar, SPECT} denotes the SPECT-based calibration factor in units of Bq/cps.

In the absence of specific binding to the bone marrow or blood cells, as indicated for PSMA therapy [27, 28], the bone marrow self-absorbed dose is solely given by the activity in the extracellular fluid of the marrow tissue [11]. The activity in the extracellular fluid of the bone marrow can be derived from the activity concentration in the blood plasma (blood method), multiplied with the red marrow extracellular fluid fraction (RMECFF = 0.19) of the bone marrow [11, 29, 30]. The activity concentration in the plasma can in turn be determined from the activity concentration in the blood (\( \left[{\overset{\sim }{A}}_{\mathrm{blood}}\right] \)) and the patient-specific haematocrit (HCT), if there is no specific binding to the blood cells [11]. This yields to:

where RMBLR corresponds to the red-marrow-to-blood activity concentration ratio [11]. m denotes the bone marrow (BM) or whole-body (WB) masses (m_{BM/WB, phantom/patient}) of either the phantom or of the patient [11, 26].

For Lu-177-Octreotate therapy, it holds that:

[11, 13]. To scale the male and female S values to the patient anatomy, an exponent of a = 1.001 and a = 0.992 was proposed for Lu-177-PSMA-617 and Lu-177-Octreotate therapy, respectively [26]. To derive the patient-specific blood TAC, 1 ml of blood of each sample was pipetted into a test tube and measured within a Cobra Gamma Counter (Packard Instrument Company, Inc., Meriden, CT), which has been previously calibrated via five 1-ml test samples of known activity concentration. For the calculation of the time-integrated blood activity concentration, a bi-exponential model was fitted to the time-activity data, followed by integration from zero to infinity according to Eq. (2).

Via subtraction of the time-integrated activity in the extracellular fluid and the time-integrated activity of the main accumulating organs from the whole-body, the respective ROB time-integrated activity (\( {\overset{\sim }{A}}_{\mathrm{ROB}} \)) was determined. The whole-body and organ time-integrated activities, \( {\overset{\sim }{A}}_{\mathrm{WB}} \) and \( {\overset{\sim }{A}}_{\mathrm{organ}} \), were determined from a mono-exponential fit to the three measurement points at 24, 48, and 72 h post-injection. All organ activities were derived from the sequential SPECT images, while for the determination of the whole-body activity, the sequential whole-body planar images were used. The kidneys were considered as main accumulating organs for both Lu-177-Octreotate and Lu-177-PSMA-617 therapy, according to the previous studies assessing dosimetric estimates [5, 7‐9]. The patient-specific volumes of interest (VOIs) for the kidneys were defined based on a percent isocontour of the organ maximum and of the quantitative SPECT at 24 h p. i. (PMOD Version 3.609), since images taken at early time points offer a high signal-to-background ratio for organ delineation. We adjusted the isocontour level for each patient in the best way with the usage of the CT as guidance. For all patients, an isocontour level of 30–40% was found to be appropriate. All kidney VOIs were copied to the following SPECT scans 48 and 72 h p. i., which were co-registered onto the SPECT scan 24 h p. i. in advance. We manually re-positioned, i.e. shifted or rotated, the kidney VOIs in case of imperfect co-registration of the individual SPECT time points. For Lu-177-Octreotate therapy, the liver and spleen were additionally included in the bone marrow absorbed dose from the organs [5]. For the patient-wise delineation of the liver and spleen, a similar approach as for the kidney definition was chosen using a 10 to 15% isocontour for the liver and a 30 to 40% isocontour for the spleen. The lower isocontour for liver delineation can be explained by the fact that NET patients often exhibit liver metastases, which lead to a heterogeneous activity accumulation with multiple hot spots.

The bone marrow absorbed dose from the ROB is finally given by the following formula according to Hindorf et al. with adjusted exponents as proposed by Traino et al. [11, 26]:

Eq. (4.2) considered all phantom- and patient-specific whole-body, ROB, bone marrow, and organ masses m_{WB/ROB/BM/organ, phantom/patient} for S value scaling. For male and female patients, b = 0.896 and b = 0.894 as well as c = 0.963 and c = 0.970 were used, as proposed by Traino et al. [26]. The bone marrow absorbed dose contribution of each individual organ is given by:

For the reference protocol (RP), the bone marrow absorbed dose from the ROB is determined from all three available whole-body planar scans (Fig. 2). For the total bone marrow absorbed dose, the absorbed dose from the three constituents, organs, blood, and ROB, was summed. For each dose constituent, the percentage contribution (PC_constituent) to the total bone marrow absorbed dose was calculated:

The proposed hybrid protocol (HP) uses a single whole-body image and sequential single-bed quantitative SPECT acquisitions of the abdomen to determine the ROB TAC, instead of deriving the ROB TAC from sequential whole-body planar imaging.

First, the abdominal effective decay constant λ_SPECT was derived via a mono-exponential fit to the total activity in the SPECT scans 24, 48, and 72 h post-therapy. Especially, all organs and all tumours were included in the fitting of the TAC, as it was the case for the determination of \( {\overset{\sim }{A}}_{\mathrm{WB}} \) from the reference protocol. This effective decay constant λ_SPECT serves as a surrogate for the reference-protocol-based whole-body effective decay constant (Fig. 2). The mono-exponential SPECT-based abdominal TAC was then scaled with a chosen base point. This base point is defined via the whole-body activity A_WB(t^∗) of a single whole-body planar image acquired at an arbitrary time point t^∗∈ 24, 48, or 72 h post-therapy. The resulting pseudo-whole-body TAC A_{WB, pseudo}(t) is intended to serve as an estimate of the reference-protocol-based whole-body TAC (Eq. (7.1)) and can be further used to determine a pseudo-whole-body time-integrated activity\( {\overset{\sim }{A}}_{\mathrm{WB},\mathrm{pseudo}} \) (Eq. (7.2)).

T_{1/2, SPECT} denotes the SPECT-based effective half-life.

Based on the hybrid model given in Eqs. (7.1) and (7.2), the bone marrow absorbed dose from the ROB can be estimated by Eqs. (4.1) and (4.2). In this work, we investigated a combination of the sequential abdominal SPECT with the whole-body planar images at 24, 48, or 72 h p. i., where each whole-body planar image was individually calibrated via the quantitative SPECT at the corresponding time point (Fig. 2). These different hybrid protocols were further denoted as HP24, HP48, and HP72. The agreement of the bone marrow absorbed doses from the ROB, as determined via the HP and the RP, was assessed. Therefore, the percentage deviation between absorbed dose estimates (PD_dose; Eq. (8)) was calculated, and a statistical test for correlation was performed (MATLAB Pearson correlation analysis).

Furthermore, the same analysis was performed regarding the total bone marrow absorbed dose estimates composed of all available constituents: the ROB (including tumours), the explicitly analysed organs, and the contribution of the blood activity. While the application of the hybrid protocol affects the bone marrow absorbed dose from the ROB, all other constituents were not altered.

For a mono-exponential TAC, the time-integrated activity is calculated as the product of the effective half-life T_1/2 and the y-axis intercept A₀ of the fit function:

The proposed hybrid protocol assumes that ideally, the SPECT-based abdominal effective half-life is equal to the whole-body effective half-life. However, in reality, differences in both half-lives will lead to deviations in the area under the whole-body TACs derived from the reference protocol and hybrid protocol, and thus in the respective whole-body and ROB time-integrated activities. Simultaneously, these deviations in the course of the TACs may affect the y-axis intercepts of the reference-protocol-based and hybrid-protocol-based TACs. To address this issue, both fit function parameters, effective half-life and the y-axis intercept, were compared for the reference protocol, HP24, HP48, and HP72. For a perfect agreement between the reference-protocol-based and hybrid-protocol-based ROB time-integrated activities, \( {\overset{\sim }{A}}_{\mathrm{RP}} \) and \( {\overset{\sim }{A}}_{\mathrm{HP}} \), the product of the ratio of reference-to-hybrid effective half-lives \( \left(\frac{T_{\mathbf{1}/\mathbf{2},\mathrm{RP}}}{T_{\mathbf{1}/\mathbf{2},\mathrm{HP}}}\right) \) and the ratio of reference-to-hybrid y-axis intercepts \( \left(\frac{A_{0,\mathrm{RP}}}{A_{0,\mathrm{HP}}}\right) \) has to yield 1: