Post-treatment image-based dosimetry

The goal of image-based personalized dosimetry calculations is to determine for each particular patient the dose (or dose-distribution) absorbed by his/her tumor(s) and organs at risk. For ¹⁸⁸Re-AHDD-Lipiodol radioembolization, the organs at risk include the normal liver, lungs and bone marrow. In addition, other organs that might uptake ¹⁸⁸Re-AHDD-Lipiodol are the stomach, kidneys, spleen, thyroid and salivary glands (see Fig. 1).

Whole-body planar images (anterior views) of three patients (no. 11, 7b, and 6) who received an intra-hepatic injection of ¹⁸⁸Re-HDD-Lipiodol. The images were acquired at 3 h, 24 h, and 48 h post-administration of the radiotracer. The quantity I_max represents the maximum pixel intensity of each planar image

In order to calculate the absorbed dose, the hybrid planar/SPECT imaging protocol was applied in our study [34, 35]. It consisted of 2–3 WB planar scans at t₁ = 3 ± 1 h, t₂ = 24 ± 1 h, and t₃ = 48 ± 4 h post-administration of ¹⁸⁸Re-AHDD-Lipiodol and one SPECT/CT of the thorax/abdominal area at t₁ = 3 ± 1 h. The WB images were used to determine the relative change of activity over time in regions of interest (i.e., the shape of the time-activity curves; TACs) while, the single SPECT/CT scan was used to determine the absolute TACs.

Secondly, the absolute TACs must be integrated over time to calculate the time-integrated activity (or total number of decays of ¹⁸⁸Re) for each tumor and organ at risk. In our study, since only 2–3 data-points were available, the time-activity data were fit to mono-exponential functions. Details of the fit and the integration of TACs are described in the following sections.

Third, the time-integrated activity of each region of interest was divided by the injected activity to yield the TIACs. This quantity allowed us to compare the accumulation of ¹⁸⁸Re-AHDD-Lipiodol in tumor/organs between different patients.

Finally, the TIACs were multiplied by the tumor and organ S-factors to determine the average absorbed dose (or dose per injected activity) in regions of interest.

A SymbiaT SPECT/CT camera (Siemens Medical, Germany) equipped with a high-energy collimator was used in all the scans. All the image-processing tasks described in this section were performed using QDOSE, the software for dosimetry calculations in radionuclide therapies (ABX-CRO, Germany). The dose calculations were performed with OLINDA/EXM 1.1 [36].

In order to acquire the planar image of the whole-body, the speed of the patient-bed was set to 15 cm/min. The WB data at both anterior and posterior views were collected using a photopeak window centered at 155 keV (20% width) and were processed according to the MIRD pamphlet No. 16 [37]. Since no transmission scans were available, attenuation and/or scatter corrections were not applied to these images. For this reason, the 2-dimensional planar data were only used to determine the relative TACs.

In order to determine the time-activity curves of the source region r_S, rough boundaries around these regions (organs/tumor) were first delineated on the WB planar images corresponding to the first time-point t₁. This was done in order to separate these organs and tumors from the other regions containing activity. Subsequently, within these rough boundaries, a 40% threshold was applied to determine the count-rate within each segmented region of interest (ROI) in both the anterior and posterior views of the patient and the geometric mean was calculated. Then, the WB planar images acquired at t₂ and t₃ were co-registered to the first WB image at t₁ using a rigid-registration algorithm and the segmented ROIs from the first image were transferred to the WB images that were acquired at the second and third time-points. As a result, the count-rate as a function of time was obtained for each source region r_S. Since the count-rate is proportional to the activity for a given patient, we can assume that the relative change in count-rate equals the relative change of activity over time in the absence of dead-time losses. The count-rate losses in the WB images due to dead-time were corrected using our previously reported method [19]. For each source region, the relative time-activity data (A_rel(r_S, t)) were fit to a mono-exponential function using the weighted least-squares method:

where a and T_eff are the parameters of the fit. The quantity T_eff represents the effective half-life of ¹⁸⁸Re-AHDD-Lipiodol in the source region r_S. It is related to the physical half-life of ¹⁸⁸Re (T_phys = 17 h) and the biological half-life of AHDD-Lipiodol (T_bio) by the following formula:

During SPECT acquisitions, a total of 32 projections (20 s/view, 16 projections/detector) were acquired over 360° around the patient (angular step = 360/32 = 11.25°). The detector field-of-view covered the lungs and liver region in one bed position. The tomographic data were collected using a photopeak window centered at 155 keV (20% width) and two narrow secondary windows placed below and above the photopeak (5% width), respectively, to be used for triple-energy window (TEW) scatter correction. In addition, a low-dose CT (kVp = 130 kV, Exposure = 47 mAs) was acquired to be used for attenuation correction. The details of each patient’s protocol (patient information, administered activity, tumor volume, and imaging times) are reported in Table 1. Please note that for 4 patients (out of 13 patients) the imaging data were acquired at only two time-points (at t₁ and t₂).

Patient information. The reported administered activities correspond to the dose-calibrator readings corrected for Lipiodol attenuation. The liver volumes were derived from pre-treatment contrast-enhanced CT images except for patients 3 and 8, which were derived from post-treatment SPECT/CT images

*Tomographic data in these patients were acquired without scatter windows. In these cases, the scatter was compensated using the attenuation map with broad-beam attenuation coefficients

Additionally, since the projection data for four patients’ tomographic studies (no. 1, 2, 3, and 8) did not include scatter windows, in these cases, the scatter correction was performed using the approximated approach with broad-beam attenuation coefficients. Also please note that one of the patients (patient no. 7) was treated twice.

The patient’s tomographic data were reconstructed using our in-house MIRG Software for quantitative ¹⁸⁸Re SPECT, as described in our previous publication [19]. The protocol consisted of standard ordered-subset expectation maximization (OSEM [38]) reconstruction (8 subsets, 12 iterations) with corrections for attenuation (CT-based), scatter (TEW [39]), dead-time (based on the correction curve determined using phantom experiments [19, 40]), and resolution recovery [41]. The estimated dead-time losses in the patient projection data were always less than 6%.

The counts in the reconstructed images were converted into units of activity by applying a camera calibration factor determined from a planar scan of a point-source, following a procedure previously described [19].

To determine the absolute TACs (A(r_S, t)), the source’s absolute activities were determined from the SPECT images (A(r_S, t₁)) and the relative TACs (A_rel(r_S, t)) were re-scaled according to these absolute source activities:

where the parameters a and T_eff are from Eq. 1, and the quantity A_rel(r_S, t₁) is obtained by evaluating Eq. 1 at t₁.

In order to calculate the absolute activity in the organs of interest, their boundaries were manually delineated based on their physical size from CT images. This segmented CT-volume was applied to the SPECT image and the activity within the volume was integrated. The boundaries of tumors, however, were obtained by a different method. Firstly, a rough boundary was manually drawn (on the CT) covering the liver segment or lobe that contained the tumor (as reported by the physicians). Secondly, a fixed threshold was applied within this boundary region on the SPECT image such that the recovered volume would equal the reported tumor volume of Table 1.

The last step to determine the time-integrated activity A~rSTD in the source region r_S involved the following calculation:

where T_D is the dose-integration period, which was equal to infinity in our study. The following assumptions were made about the TAC in time-intervals where no imaging data were available: for t = 0 to t₁, the source-organ activity was assumed to grow linearly from 0 to A(r_S, t₁), and for t > t₃, the activity was extrapolated assuming a mono-exponential clearance following the ¹⁸⁸Re physical decay. The time-integrated activity A~rSTD was then calculated by analytical integration of this TAC. The resulting time-integrated activity in each source organ was divided by the injected activity (A₀) to determine the TIAC a~rSTD:

The absorbed radiation doses in organs of interest were calculated using OLINDA by combining the estimated TIACs with the pre-calculated organ-level S-factors of an adult male/female human phantom. The S-factors for each organ were adjusted to the organ mass of each patient obtained from CT images. The doses absorbed by tumors were obtained by multiplying the tumors’ TIACs by the ‘spherical model’ S-factors (adjusted for each tumor volume) available in OLINDA. Information and tabulated S-values for organs at risk and spheres of different sizes is included in the Additional file 1.

In this work, we determined pharmacokinetics and dosimetry for the regions which were visible within the field-of-view of SPECT/CT images, as only for them the activity could be accurately quantified. These regions included the tumor, entire liver, lungs, stomach, spleen, and kidneys. Please note that the “remainder of the body” TIAC term in OLINDA, which mostly corresponded to activity in salivary/parotid glands, thyroid, and urinary bladder (see the “Biodistribution of 188Re-AHDD-Lipiodol” section) was not determined in our study. The dose to bone marrow was not estimated due to the limited accuracy of image-based methods to determine it. An alternative method to estimate bone marrow dose in ¹⁸⁸Re-Lipiodol studies based on blood-sample counting was described in Zanzonico and Divgi 2008 [29].