Acquisition count rates and data selection
We first examined the observed wide-spectrum count rate per projection for the phantom (20 acquisitions, Fig. 1a and b) and patients (four representative acquisitions, Fig. 1c). In dual-detector mode, the Intevo used for phantom studies saturated at 268 kcps per detector when both detectors were simultaneously exposed to the highest activities (Fig. 1a). Because of the asymmetrical activity distribution in the phantom, it was still possible to obtain non-saturated acquisition data up to 9.34 GBq (i.e., up to 400 kcps on detector 1). When acquiring in single-detector mode (i.e., using only detector 1 over 360°), the absolute saturation level increased to 592 kcps, allowing to quantify up to 15.34 GBq (Fig. 1b). We excluded the two phantom datasets with the lowest activity (0.16 and 0.21 GBq), for which the quantification inaccuracy exceeded 5% in whole phantom or hot objects (Fig. 2). We thus included ten dual-detector acquisitions from 0.32 to 9.34 GBq, plus six single-detector acquisitions from 9.36 to 15.33 GBq for further analysis (n = 16).
For the Symbia on which the patient acquisitions were performed, the differential per-detector absolute saturation threshold that we previously observed (7, 8) was obvious in two patients on Day 1 after treatment (patients 13 and 14, the latter case illustrated in Fig. 1c). These two acquisitions were excluded from further analysis, resulting in 26 quantifiable acquisitions and 12 valid two-timepoint dosimetry studies.
Calibration factors and dead-time constant
The calibration factor and dead-time constant calculated using the valid reconstructed SPECT data (n = 16) and previously described methods (8) were 10.05 ± 0.04 cps/MBq and 0.56 ± 0.14 µs, respectively, for the Intevo. For the Symbia, we used those previously determined by Frezza et al. (8): 9.36 ± 0.01 cps/MBq and 0.550 ± 0.003 µs, respectively. Considering the close similarity of the dead-time constants and the larger uncertainty of that of the Intevo, we elected to use 0.55 µs across the two systems to generate a common DTCF look-up table for both.
Validation of pre-reconstruction dead-time correction
To assess the impact of the random component involved in the methods by which the lost counts were injected into projections, DTCM1 and DTCM2 processing and subsequent reconstructions were done in triplicate for each dataset. For all VOIs under study, both in phantom and in patients, and for both DTCM1 and DTCM2, the median and the maximum coefficients of variation of the VOI counts were 0.27% and 1.50%, respectively. This maximum coefficient of variation was found for tumours in patients (n = 118), whereas those for all other VOIs did not exceed 0.81%. On this basis, we considered that the randomness with which fractional lost counts are distributed in projections before reconstruction has a negligible impact on quantification.
In principle, when using the average DTCF of the acquisition (i.e., based on the average wide-spectrum count rate from all projections), adding lost counts before (i.e., DTCM2) or after (i.e., DTCM3) the reconstruction should be equivalent. For all VOIs under study, both in the phantom and in patients, the median and maximum relative differences in VOI counts between DTCM2 and DTCM3 were 0.06% and − 0.84%, respectively. We thus concluded that DTCM2 is indeed equivalent to DTCM3, and since it offers no advantage over the latter, it was not studied further.
Per-volume vs. per-projection dead-time correction
For the phantom, on a per-VOI basis, the largest median and maximum relative count differences between DTCM3 and DTCM1 were found for the 200-mL background VOI, at 1.08% and 2.38%, respectively. These values were smaller for the other VOIs: -0.28% and − 1.49% for the large saline bag, 0.16% and 0.40% for the small saline bag, and − 0.18% and − 1.24% for the whole phantom (Fig. 3a). There was a trend towards increasing differences as the count rate increased for most VOIs. Two selected slices of the phantom corrected with DTCM1 and DTCM3 are represented along with parametric images of the per-voxel relative count difference (Fig. 4). Although the voxel-to-voxel differences appear heterogeneously distributed, they are of relatively small amplitude.
In patients, on a per-VOI basis, the largest count difference between DTCM3 vs. DTCM1 did not exceed 4% for any VOIs, except for the bone marrow, a VOI with low signal, for which the difference reached − 6.37% in only one case (Fig. 3b). The few occurrences of differences larger than 2% were found at higher counting rates when the DTCF was superior to 1.1.
Impact of dead-time correction methods on dosimetry
In twelve patients with a valid 2-timepoint SPECT/CT study, the median (range) self-absorbed doses were 1.29 (0.23–3.62) Gy for the bone marrow, 6.11 (4.01–9.79) Gy for the left kidney, 5.18 Gy (2.36–7.04) for the right kidney, 5.98 Gy (3.19–7.80) for both kidneys averaged, and 39.76 Gy (1.13–155.68, n = 54) for tumours. The largest median and maximum relative dose differences between DTCM1 vs. DTCM3 were 0.31% and 3.24%, respectively (Fig. 5). Of 102 datapoints, only nine exceeded ± 1%. These were more frequent when DTCF was superior to 1.1.