Feasibility of simplifying renal dosimetry in ¹⁷⁷Lu peptide receptor radionuclide therapy

Peptide receptor radionuclide therapy (PRRT), an established treatment for neuroendocrine tumours (NET), has recently received regulatory approval based on the results from the randomized phase III trial NETTER-1 [1]. Further optimization of PRRT may be achieved not only through improved patient selection with the use of clinical criteria [2], imaging [3] and molecular analyses[4] but also through improving the way the treatment is planned and carried through. The optimization strategies differ from centre to centre and include, but are not limited to, combining PRRT with chemotherapy [5, 6], combining different radionuclides [7] and using dosimetry to personalize treatment [8], the latter being the method of choice by our groups.

In external beam radiotherapy and brachytherapy, it is standard procedure to individually plan treatment by calculating absorbed dose (AD) to both target (i.e. tumour) and organs at risk, with the aim of optimizing the AD delivery so as to achieve a high probability of anti-tumour effect with a low risk as reasonably achievable of serious toxicity. This concept can also be applied to systemic radiotherapy such as PRRT, although there is as of yet no international consensus on how this should be done.

Our groups have, over the past several years, systematically worked to improve methods and clinical applications in the realm of dosimetry-guided systemic radiotherapy, with the long-term aim of improving treatment planning and results [9–17]. As part of this development, we designed a phase II trial (NCT01456078, “Iluminet”) to investigate the feasibility, safety and efficacy of individualized PRRT based on patient selection and dosimetry. The trial has been approved by the institutional ethics committee (ref. no. 2011/287) and national regulatory authorities. Further details on the protocol can be found at www.clinicaltrials.gov. All patients have given their written informed consent to participate.

In the Iluminet trial, the hypothesis was that by using individualized treatment planning, the balance between treatment effect and toxicity is optimized. Patients with advanced, progressive NET were treated with 7400 MBq ¹⁷⁷Lu-dotatate at 8–12-week intervals, and detailed post-therapeutic imaging and dosimetry were performed in all patients after every cycle. The number of cycles each patient received was determined by a combined evaluation of dosimetry, treatment effect and toxicity. Patients with risk factors for nephrotoxicity received treatment up to a cumulative renal biologically effective dose (BED) of (27 ± 2) Gy, and patients without such risk factors were offered to continue up to (40 ± 2) Gy. This individualized approach led to a wide range in the number of cycles per patient, differing substantially from the standard four treatment cycles [18].

The dosimetric method used per protocol in the Iluminet trial is a hybrid planar-SPECT/CT method, where the time-retention curve is derived from serial planar images, and the absorbed dose-rate information from a SPECT/CT is used for rescaling into a time-dose rate curve. Performing such dosimetry for each cycle is time-consuming both for the patients and for the therapeutic team of technologists, physicists and physicians. In the context of a prospective clinical trial exploring the feasibility of individualized, dosimetry-based treatment planning, it is justified to use such a complex method. However, if this concept is to be brought to clinical routine, it is of interest to determine whether simplifications can be made without substantially affecting the estimated ADs.

In the present study, we use the image data acquired during the trial and define a set of alternative treatment planning strategies, each representing a simplification in terms of image acquisition and dosimetric calculations. The alternative strategies include the current standard of 4 cycles to all patients, planar dosimetry, simplified hybrid dosimetry and dosimetry based on one SPECT/CT only. Additionally, we have applied the method proposed by Hänscheid et al. [19] of using one single measurement after 4 days for both planar and SPECT/CT data. The results from the simplified strategies are compared to the results from the protocol-prescribed hybrid planar-SPECT/CT-based method and are evaluated both in terms of per-cycle differences as well as the cumulative absorbed dose over all cycles.

Thus, the aim of the work presented in this paper is to investigate the ADs that are obtained when simplified treatment strategies or dosimetry schemes are used as alternatives to the more complex protocol-prescribed method, in order to make conclusions of possible simplifications for future dosimetry-based ¹⁷⁷Lu-PRRT.

The image data on which the present analyses are based include the same patients as in the previously published interim analysis of the Iluminet trial [18]. Of the 51 patients that had been included at the time of the interim analysis, 22 had completed treatment as planned, i.e. had reached the targeted BED to the kidneys. It is on this subset of patients that the present analyses have been performed, comprising dosimetry of 119 cycles in total.

Quantification of activity, AD and BED

The procedure for image-based dosimetry is summarized in Fig. 1.

The dosimetry calculations were performed using a program package called LundADose, implemented in the IDL programming language (Harris Geospatial Solutions, Broomfield, CO, USA). Prior to the actual computations, patient images were exported from the respective gamma camera systems, in DICOM image format, to the Interfile 3.3 format used internally in LundADose.

The planar images were processed using a pixel-based implementation of the conjugate-view method, as previously reported [21, 22]. In the processed whole-body images from the four time points, regions of interest (ROIs) over the left and right kidneys were delineated, as well as a background ROI below each kidney (Fig. 2). Since the main purpose of the planar-based values was to determine the shape of the renal time-activity curve rather than the total kidney-activity content, it was not necessary that the ROIs encompassed the whole kidneys. Instead, care was taken to avoid overlapping tissues such as the liver or tumours. In cases when one kidney was severely overlapped by extra-renal activity uptakes, the effective half-time from the least overlapped kidney (often left) was used as an estimate also for the contralateral kidney. Background correction was accomplished by determining the activity per pixel in the respective background ROIs and then scaling to the number of pixels in the kidney ROIs, as well as to an estimated thickness of the background compartment within the kidney ROI [22].

Voxel-wise SPECT activity quantification was achieved as part of an iterative ordered-subsets expectation maximization image reconstruction, including corrections for attenuation and scatter, using a model-based method [23] and collimator response, and employing eight iterations and ten subsets. In order to use the CT image for attenuation and scatter corrections, the CT image values were first converted to mass density using a calibration-phantom-based relationship, as previously described [21]. From the density map, an attenuation map was generated by multiplication to mass attenuation coefficients valid for the soft tissue and cortical bone, for 208 keV. The quantitative SPECT image and the CT-derived mass-density image were used as basis for a Monte Carlo voxel-based dose-rate calculation using the EGS4/PRESTA algorithm [24]. Volumes of interest (VOIs) were drawn, by specialized technologists, along the kidney contour in the SPECT/CT images, mainly by guidance from the transversal CT. The VOIs were applied to the 3D absorbed-dose rate images, and the median absorbed-dose rate in the VOI was determined. This value was then divided by a recovery factor of 0.85, which had been established in a previous study where manual segmentation was performed by several operators in Monte Carlo-simulated SPECT/CT images [25]. This value of 0.85 was thus considered valid for all patient kidneys using the chosen methodology for image segmentation and SPECT image reconstruction.

A series of time-dose rate values at different times, R(t), were derived by combining the SPECT-derived absorbed-dose rate acquired at t_S, i.e. R(t_S), with the planar-derived activity values from different time points, A_P(t), according to

$$ R(t)=\frac{A_{\mathrm{P}}(t)}{A_{\mathrm{P}}\left({t}_{\mathrm{S}}\right)}\cdotp R\left({t}_{\mathrm{S}}\right) $$

(1)

A curve, consisting of an initial straight line followed by a trailing mono-exponential function, was fitted to time-dose rate values. The mono-exponential function was fitted to the three last data points using weighted least squares, and the initial line was then determined analytically based on the values of the mono-exponential function at the time of the second data point and the first data point. This curve shape, in particular the initial straight line (equivalent to a trapezoid integration), was used as the simplest approach, realizing that the true, underlying renal time-activity curve, including the very fast initial redistribution [16, 26], was not resolvable from the two data points acquired during the first 24 h. The AD was determined as the area under the curve, calculated by analytical integration. Finally, the absorbed-dose rate curve was used as input to the calculation of the BED using a convolution-based method [12] with radiobiologic parameters α/β of 2.6 Gy and a repair half-time of 2.8 h [27].

The decision of whether the patient was to receive further treatment cycles was based on an overall evaluation of the cumulated renal BED, the projected renal BED in case of further treatment in relation to the target BED (27 or 40 Gy) as well as tumour effect and toxicity of treatment thus far.

The overall aim in this study was to elucidate whether it is possible to use a less demanding treatment strategy or dosimetry method in individualized ¹⁷⁷Lu-dotatate therapy, as compared to that used in the ongoing clinical trial (Iluminet), while maintaining a similar degree of accuracy in renal AD estimates.

For the clinical trial protocol, a hybrid dosimetry method has been used. This choice was a compromise between the axial coverage offered by the whole-body planar images, and the superior accuracy of SPECT/CT. Moreover, at the time of the writing of the clinical trial protocol, performing serial SPECTs in each cycle, including the subsequent processing and analysis of 3D images, was deemed too time-consuming to be feasible.

For the evaluation of the AD per cycle for the alternative treatment strategies (Fig. 3), we conclude that the dosimetry methods in which SPECT/CT are included are superior to the purely planar-based methods and that including one SPECT/CT in each cycle improves the dosimetric accuracy considerably. Results based on the cumulative AD (Table 2) are consistent with results from the per-cycle analysis. Dosimetry method G, including one SPECT/CT in each cycle at 96 h, yields results that are equivalent with the protocol dosimetry method P but with much lower demands on resources. Methods E and F, which also use SPECT in every cycle, give similar but slightly less consistent results.

Method G was implemented following the original work by Hänscheid et al. [19], where an approximate method for estimation of the area under the time-activity curve based on one single image acquisition was developed. The approximation can be derived from the expression of the absorbed dose for a mono-exponential curve, following

$$ D=\kern0.5em \frac{R\left({t}_{\mathrm{ref}}\right)\cdotp \exp \left(\lambda\ {t}_{\mathrm{ref}}\right)}{\lambda }=\frac{R\left({t}_{\mathrm{ref}}\right)\cdotp {2}^{\kern0.5em \left(\frac{t_{\mathrm{ref}}}{T_{1/2}}\right)}\cdotp {T}_{1/2}}{\ln (2)}\approx \frac{R\left({t}_{\mathrm{ref}}\right)\cdotp 2\cdotp {t}_{\mathrm{ref}}}{\ln (2)} $$

(9)

where the last, approximate equality is exact when the imaging time is equal to the effective half-time (t_ref = T_1/2) or two times the effective half-time (t_ref = 2 T_1/2). Hänscheid et al. [19] noted that the right-hand expression yields valid approximations when 0.75 T_1/2 < t_ref < 2.5 T_1/2 and in particular when t_ref = 96 h. If we denote the right-hand term as \( \widehat{D}, \) the theoretical fractional error, E₁, becomes

$$ {E}_1=\frac{\left(\widehat{D}-D\right)}{D}=2f\cdotp {2}^{-f}-1 $$

(10)

with f = t_ref/T_1/2. Figure 4 shows the theoretical fractional error, E₁, as a function of t_ref for an effective half-time 51.6 h, as obtained in a previous study [18]. It was considered of interest to evaluate the fractional error for patient data for different choices of t_ref. Moreover, since in practice the imaging time point cannot be tuned to the effective half-life for a specific patient and cycle, for a given choice of t_ref, the dispersion in effective half-times will translate into a variance in the fractional error. The data underlying results in Fig. 3 (method G) were thus reanalysed to a patient-based fractional error, E₂ = (D_G − D_P)/D_P , where D_G and D_P are the absorbed doses obtained from method G and P, respectively. Figure 4 shows the mean value of E₂ obtained over all patients and cycles, the SD around the mean and the root-mean square deviation (RMSD) in E₂, when the value of t_ref in Eq. 8 is varied between 24 and 144 h. It appears that the theoretical fractional error (E₁) compares well with the fractional error obtained for patients (mean of E₂) and that the overall deviation, as described by the RMSD, contains both systematic (mean of E₂) and random components (SD of E₂). When the dispersion in effective half-times is comparably modest, as for the kidney data analysed herein, the RMSD exhibits a valley when T_1/2 < t_ref < 2T_1/2 and is lowest near 96 h (t_ref ≈ 2T_1/2). For an imaging time near 50 h (t_ref ≈ T_1/2), the systematic component of the error is equally low as for 96 h, while the SD is higher, thus yielding a slightly higher RMSD. The RMSD is also more sensitive to the exact acquisition time at 50 h than at 96 h. Thus, acquisition at approximately 4 days appears to be a valid choice for this approximation. This is also confirmed by the uncertainty analysis based on SPECT/CT on day 4 (Additional file 1), where method G yields ADs that are equivalent to the protocol results. One drawback with our evaluation of method G is that the effective half-time estimated for the individual patient is included in the calculation of the absorbed-dose rate at 96 h since our SPECT/CTs were not acquired at 96 h. However, we still find it motivated to perform an independent evaluation of this new method, and the included implementation is deemed to be the fairest.

The comparably small deviations obtained using method F, i.e. one SPECT/CT for each cycle and a standard effective half-time for all patients, were unexpected and prompted further investigation. Figure 5 shows data derived from a larger patient material (80 subjects) than the one used for the analyses above. From this figure, we see that the shape of the time-activity curves after 24 h (A and B) as described by the effective half-time (C) is similar between patients, the median being 51.7 h. It is possible, however, that this uniformity in effective half-time is partly due to the uniformity of patient selection and procedures dictated by the clinical trial protocol. It would be of interest to perform a similar analysis in a real-life setting and in different PRRT centres. The vast spread in the first data point in the time-activity curve is a result of the imaging being performed at a time when there is a fast initial turnover and washout of ¹⁷⁷Lu-dotatate via the kidneys [26]. This thus supports the choice of fitting function where the first data point, acquired at approximately 1 h, is decoupled from the fitting of the exponential tail and is thus not allowed to influence the assessment of the effective half-time. In spite of the vast spread observed, the AD using method F is in most cases consistent with the protocol dosimetry method P. This is explained by results in Fig. 5d where it is observed that the majority of the AD is delivered after the first 24 h (left kidney median 76% (interquartile range 73 and 79%), right kidney median 75% (interquartile range 72 and 79%)).

The conversion from activity to absorbed-dose rate is, according to the clinical protocol, performed using Monte Carlo-based radiation-transport calculation in a voxelized geometry. Since the range of the electrons emitted from ¹⁷⁷Lu is shorter than the spatial resolution of the SPECT images, the radiation-transport calculation could possibly have been simplified to a multiplication by the assumption that the electron kinetic energy is absorbed locally in the voxel [24]. However, it was considered important not to neglect the photon contribution, since parts of the kidneys are located near organs with high uptake such as the spleen, and since in addition to gamma radiation, there are also low-energy X-rays emitted in the ¹⁷⁷Lu decay that contribute to the self-absorbed dose. The Monte Carlo method was considered the best choice based on accuracy and availability. Notably, given that only one SPECT/CT is available, it is assumed that the fractional contribution from photons to the total absorbed energy in the kidneys is constant during the entire treatment cycle, whereas in reality, the photon contribution varies with time depending on the activity present in surrounding tissues.

When attempting a conclusion from these results, we can look at them from two different angles. From a methodological point of view, we want the most accurate method to be used, while from a clinical perspective, the question is how much accuracy is really needed in relation to the observed efficacy and toxicity. Of the methods compared in this analysis, dosimetry as per protocol is presumably the most accurate one (as the rest are simplifications of the same), but doing four SPECTs/CT in each cycle would probably yield even more accurate results although to the price of a more time-consuming procedure. Among the simplified methods analysed, we would choose the one with the smallest mean difference for each cycle and the narrowest LOAs.

The methods which best fit these criteria are E and G, both of which incorporate SPECT-based dosimetry in every cycle but are less resource-intense than the reference method P. The strategy that deviates the most is unequivocally method A, i.e. giving 4 cycles to all patients, followed by the planar-based dosimetry methods (B, C and I). This is consistent with what is already known about the drawbacks of planar dosimetry, namely the difficulties in taking into account individual variations in activity concentration, especially in tumours, the liver and intestines, which may then lead to over- or underestimation of the renal AD.

When choosing between methods E and G, one aspect to take into account is that for most PRRT centres, it would be more resource-consuming to perform a SPECT at 96 h since the patients would have to come back to the hospital. Whether or not to still go for method G then becomes a matter of whether or not the small difference seen in the Bland-Altman analyses (2SD of 11 and 17% for methods G and E, respectively) makes a relevant difference in real-life patient management.

From a clinical perspective, on the other hand, some would argue that to give 4 cycles to all patients is already a very effective treatment with a low degree of toxicity, so the need for further optimization is marginal. There are two objections to this: firstly, our obligation to know what we are doing when employing radiation for therapeutic purposes (analogous to the demands in other radiation therapy modalities), and secondly, when these patients progress months or years after their 4 cycles of PRRT, we would like to retreat them since the majority of the cases have not reached the limits for the organs at risk. Without initial dosimetric estimates, we have no grounds on which to plan further treatment.

Given the highly clinically and statistically significant effect of PRRT demonstrated in the NETTER-1 trial [1], together with the ample experience and long safety follow-up with the standard 4 cycles of treatment, perhaps the optimal treatment strategy is a mix between this and dosimetry-based treatment. This could be achieved by initially giving 4 cycles to all (or less if tolerance limits are reached before), but performing dosimetry as well. When the tumour later progresses, the patient can be re-treated to risk organ tolerance limits. In this manner, we get a good risk-benefit balance at each stage of the disease—early in the disease when the patient has a longer expected survival, we use a low-risk treatment, but once progression has been confirmed and the prognosis is another, it is more reasonable to assume the possible risks of higher ADs to risk organs associated with further treatment. Even with the relatively high renal BED limits used in the Iluminet trial, the toxicity has so far been limited (unpublished data), so it is still an open question which limits we should use in the future. Perhaps further retreatment beyond the limits used in this trial will be feasible for some patients under certain conditions. For this reason also, it will be of essence to incorporate dosimetry in PRRT planning, both in clinical routine and future clinical trials, and thereby progressively increase our understanding and body of knowledge.

From a methodological and clinical point of view, we need an accurate dosimetric method demanding a reasonable amount of resources. Dosimetry based on one SPECT/cycle complies with these requirements, and it seems that performing the SPECT at 96 h gives reliable results.