Clinical implementation of PLANET® Dose for dosimetric assessment after [¹⁷⁷Lu]Lu-DOTA-TATE: comparison with Dosimetry Toolkit® and OLINDA/EXM® V1.0

Dosimetry applications are expanding and the number of nuclear medicine departments performing patient dosimetry is growing, especially for patients with neuroendocrine tumours receiving peptide receptor radionuclide therapy [1]. This is possible thanks to the multidisciplinary collaboration between nuclear medicine physicians, medical physicists, and nuclear medicine technologists [2].

In the clinic, many medical teams have developed their own methodology using the tools available in their department and according to their own organizational possibilities [3‐5]. This has led to the local implementation of in-house developed dosimetry software and programs [6‐10]. However, the legislation on Medical Devices restricts their use to clinical trials. For routine clinical use, a software package must have received the CE mark. Therefore, many medical teams are now acquiring commercial packages, because obtaining the CE label is usually beyond the missions/capabilities of academic structures and due to the increasing availability of commercial software tools. In a previous article we evaluated some commercial packages already on the market [11]. An updated table (Table 1) is presented below, but the field is rapidly and constantly evolving. Only CE-marked software tools are presented (therefore, STRATOS from the Philips research station Imalytics is not included), and some features that are still under development may not have been approved yet.

In our department, dosimetry is integrated in the clinical routine for patients with neuroendocrine tumours treated by peptide receptor radionuclide therapy with [¹⁷⁷Lu]Lu-DOTA-TATE. Up to now, dosimetry analyses of organs at risk (OAR) have been performed using the combination of Dosimetry Toolkit® (GE Healthcare, Milwaukee, USA) and OLINDA/EXM® V1.0 to calculate the Time Integrated Activity Coefficient (TIAC) and organ-level absorbed dose (AD), respectively [12]. Recently, our department acquired PLANET® Dose (DOSIsoft SA, Cachan, France), a new CE-marked commercial dosimetry workstation. The initial motivation for changing was to use a vendor-neutral solution to support multicentric trials and to put in place a central dosimetry system. However, before its clinical routine implementation, we wanted to compare this new dosimetric package with our internal reference. Our validation plan involved comparing the results obtained on phantom (calibration factor, TIAC) and in patients (mean AD to OARs, TIACs and organ volumes), using similar parameters in terms of segmentation, registration, and time activity curve fitting. This allowed assessing the consistency between the dosimetric results obtained using Dosimetry Toolkit® and OLINDA/EXM® V1.0 (our reference) and with PLANET® Dose.

Due to the need of implementing central dosimetric data processing for multicentre trials, our department acquired a vendor-neutral solution, PLANET® Dose, for dosimetry assessments. However, to take advantage of the experience and data already acquired with Dosimetry Toolkit® and OLINDA/EXM® V1.0, we needed to compare their results. Overall, the two systems gave similar dosimetry results in the phantom study (CF and TIAC) and also when using imaging data from patients who received two injections of [¹⁷⁷Lu]Lu-DOTA-TATE (mean organ masses, TIACs and ADs for liver, kidneys and spleen).

In a recent study, Huizing et al. [16] compared dosimetry results obtained with Hybrid Viewer Dosimetry and PLANET® Dose in ten patients using hybrid imaging data obtained from 0.5 h to 72 h post-injection of [¹⁷⁷Lu]Lu-DOTA-TATE. Although our study also concerned PLANET® Dose, our reference method was Dosimetry Toolkit®. Furthermore, our study was based on complete 3D imaging data from 4 to 192 h post-injection in 21 patients undergoing [¹⁷⁷Lu]Lu-DOTA-TATE treatment, thus representing a total of 40 dosimetry analysis.

In our study, the reconstruction step was performed with the GE application “Preparation for Dosimetry Toolkit” because PLANET® Dose does not include this functionality. The latter accepts reconstructed data supplied by others workstations, unlike the “Dosimetry Toolkit” application.

In a clinical dosimetry study, the preliminary step of calibration is crucial to obtain accurate activity quantification [28, 29]. A detailed calibration protocol is the cornerstone to achieve accurate and reliable image quantification in multicentric studies [30‐33]. According to the GE recommendations, CF determined by planar acquisition of a 15 cm diameter Petri dish partially filled with a solution of ¹⁷⁷Lu is sufficient for Dosimetry Toolkit®. However as our first aim was to monitor the AD to OARs, we used a large phantom (i.e. a bottle with a volume relatively close to that of kidney). The same methodology based on SPECT/CT imaging was followed for both calibration and clinical imaging. As described by Gustafsson et al. [34], SPECT image segmentation is essential to determine the activity concentration. For our phantom-based study, we selected a fixed volume threshold method based on an automatically drawn isocontour around the bottle with a volume of 200 mL. This volume was segmented on the SPECT images acquired at the first time point and rigidly copied to the others (i.e. rigid propagation). This step is directly affected by the accuracy of the registration of all SPECT/CT images using a rigid algorithm [11, 35, 36]. The obtained CF was similar (within 10%) to the one determined by Peters et al. [31] using a comparable gamma camera model.

Moreover, CFs are expressed in counts.s⁻¹.MBq⁻¹ by Dosimetry Toolkit® and in Bq.count⁻¹ by PLANET® Dose. This implies that the CF must be modified with the acquisition duration for PLANET® Dose. This also highlights a risk of errors because of the lack of a standardized method for introducing CFs in dosimetry software tools. The crucial recommendation at this step is that the conditions used in clinical studies in terms of acquisition and reconstruction parameters must be similar to those used for calibration. We observed a negligible variation of the CF values over time, from T = 0–216 h. This means that the same CF can be used for each time point. This observation is particularly interesting for Dosimetry Toolkit® in which a single CF must be entered, whereas a different CF can be used at each time point with PLANET® Dose.

In the patient study, the liver masses obtained with the two packages were more similar than those obtained for smaller organs, such as kidneys and spleen, and the variability increased with the OAR decreasing size (Table 4b). Actually, as the DICOM-RT-Structure import could not be used in Dosimetry Toolkit®, organ delineation was done manually for each dosimetry package, leading to operator-induced variability. Thus, a small difference between contours could generate a more important relative deviation in smaller than larger organs. Moreover, the results presented in Table 5a–c seem to indicate a volume/mass effect. The RMSD of the AD values increased with the OAR decreasing size. For spleen, this could be explained by our package comparison methodology that used similar parameters in terms of registration and segmentation with constant volumes over time. Indeed, when using Dosimetry Toolkit®, the volume delineated on the first image was maintained, but adjusted by translation or rotation at each successive time point, when necessary. However, this step was not available in PLANET® Dose. Therefore, to use similar approaches with both packages, in PLANET® Dose we chose to perform an organ-based registration followed by rigid propagation (i.e. the contour registered in the first CT image was exactly propagated to the images acquired at the other time points). Thus, for some patients, the initial spleen contour did not fully match the organ contour at the other time points, and included tissues with different density. This implied an important deviation when calculating the AD with density correction. For instance, in the dosimetry analysis n°31, part of the spleen contour was moved to the left lung at other time points, and the AD was overestimated because of the density correction. For kidneys, AD variations between packages could be explained by the fact that in PLANET® Dose, left and right kidney are considered separately, while in OLINDA/EXM, the two kidneys are considered as a single organ. This may affect pharmacokinetic assessments, especially when rigid registration is considered. Therefore, we tested an additional approach that uses the segmentations and TIACs provided by PLANET® Dose and the dosimetric results (ADs) obtained with PLANET® Dose LDM or DK and with OLINDA/EXM® V1.0 (Table 6). The differences between approaches were reduced when the comparison considered only the AD calculation. This is in agreement with the idea that registration and segmentation are major steps, and probably induce more variability then the AD calculation step. Grassi et al. [36] showed that ADs to organs are significantly affected by the registration algorithm used. Indeed, due to respiratory motion during SPECT/CT and organ deformation, registration errors can happen. Although this may not influence the results of our comparison, it is clear that additional studies on registration and time activity curve fitting are needed, by taking advantage of the additional possibilities available in PLANET® Dose. Moreover, for accurate registration, position reproducibility during the four SPECT/CT acquisitions is crucial. Thus, patient set-up and immobilization devices are strongly recommended.

The Bland–Altman plots showed that the biases were quite low for all organs. The limits of agreement were rather tight with a maximum value of approximately 1 Gy for spleen. Moreover, the results for liver highlighted a trend for slightly higher AD values obtained with Dosimetry Toolkit + OLINDA than PLANET® Dose when the AD increased. This trend was more pronounced for patients with liver metastases, i.e. high activity gradients within the liver. One hypothesis is that cross-absorbed doses (from photons) may be different for heterogeneous vs. homogeneous activity distributions. Another point to consider is the density (homogeneous for OLINDA vs. voxel-based and coming from the CT images for PLANET® Dose). This may influence the AD calculation, but it is not clear to which extent. This certainly deserves to be thoroughly investigated, probably using Monte Carlo modelling to take into account also the possible impact of local density corrections [37].

The concordance evaluation (Lin’s coefficient value of 0.99) highlighted an excellent agreement between methods. Moreover, the dosimetry results obtained using PLANET® Dose (AD to liver, kidneys and spleen of 0.45 ± 0.50 Gy/GBq, 0.45 ± 0.13 Gy/GBq and 0.62 ± 0.17 Gy/GBq respectively) are in agreement with those of the literature [38].

As proposed by Gear et al. [39] in a practical guidance paper, the uncertainties at each step of the dosimetry analysis should be determined to express the accuracy of the dosimetry results. Currently, PLANET® Dose allows evaluating the relative proportion of interpolation (between time points) and extrapolation (after the last time point). Regardless of the goodness of fit, this is a good indication of the relevance of time sampling, but it is not sufficient to fully characterize the uncertainty associated with the dosimetric workflow. Such study requires important developments and will be implemented in the future.

From a qualitative point of view, PLANET® Dose is a user-friendly commercial solution that proposes a wide range of tools for segmentation, and several analytic fit functions. The time necessary for a dosimetry analysis is significantly reduced. Therefore, considering the good agreement with our reference dosimetry method, the concordance of the dosimetry results with the literature, the added value of this software (easy contouring, wide choice of time activity curve fitting models, time saving), and the fact that the observed differences were explainable and clinically acceptable, we think that the PLANET® Dose software can replace our current dosimetry package without any correction for dosimetry analysis. We can now start to investigate the different methodological possibilities offered by PLANET® Dose, such as elastic registration and propagation, time integration activity with multiple exponentials, AD rate computation at the voxel level, DICOM-RT import and export of structures. Its potential can now be fully explored, particularly for the determination of the tumour AD and for investigating the AD-response correlations.

In this work we compared the dosimetric results obtained with the software currently used in our department and with a new dosimetry software package available on the market. The ADs to OARs (liver, kidney and spleen) obtained with PLANET® Dose were concordant with those calculated with GE Dosimetry Toolkit® and OLINDA/EXM® V1.0, and in agreement with the literature. These results allow us to use PLANET® Dose in clinical routine for patient dosimetry after targeted radiotherapy with [¹⁷⁷Lu]Lu-DOTA-TATE.