Fast [¹⁷⁷Lu]Lu-PSMA SPECT/CT with dual-energy acquisition and Monte Carlo-based scatter correction

Abstract

Purpose

Whole-body post-therapy SPECT/CT plays an important role in [¹⁷⁷Lu]Lu-PSMA-617 (LuPSMA) radiopharmaceutical therapy for verifying biodistribution, assessing treatment response, and performing radiation dosimetry. However, conventional dual-head NaI(Tl)-based scanners typically require lengthy scans, which can be resource-intensive and challenging for patient compliance, limiting their routine use in clinical practice. This study explores reducing acquisition duration using dual-energy peaks to increase count statistics and Monte Carlo-based scatter correction (MCS) to maintain image quality and quantitative accuracy under low-count conditions.

Methods

LuPSMA scans (33 studies from 29 patients) covering mid-thigh to vertex (3 bed positions) were acquired using a dual-head NaI(Tl)-based scanner (Siemens Intevo) with both 208 and 113 keV photopeaks and corresponding scatter windows. Full-duration 12 min/bed acquisitions were reconstructed using MCS (Hermia SPECT Reconstruction, Hermes) and the scanner’s energy-window-based scatter correction (EWS), where MCS reconstruction at 12 min/bed served as the reference standard. Shorter-duration projections (1.5, 3, and 6 min/bed) were synthetically generated and reconstructed using MCS. Image quality metrics, including bias, coefficient of variation (CV), contrast ratio (CR), and signal-to-noise ratio (SNR), were evaluated across segmented organs and tumors.

Results

MCS reconstructions at 1.5–6 min/bed showed < 6% bias in mean organ uptake (median ~ 2%) relative to the 12 min/bed MCS (reference), significantly lower (p < 0.01) than the bias for 12 min/bed EWS (median ~ 3%). For mean tumor uptake, 3–6 min/bed MCS showed < 10% bias (median ~ 5%), with no significant difference from 12 min/bed EWS (median ~ 6%). CR remained consistent across all reconstructions (median ~ 5). MCS at 12 min/bed had slightly better CV, CR, and SNR than EWS at 12 min/bed, though differences were not statistically significant. MCS at 6 min/bed was not significantly different from either full-duration reconstruction. Visual assessments supported these findings, with MCS benefits over EWS more pronounced at lower count levels.

Conclusion

These findings indicate that using dual-energy acquisition with MCS, scan times as short as 3–6 min/bed on conventional NaI(Tl)-based scanners maintain sufficient image quality for lesion detection, therapy monitoring, and dosimetry. These reductions in scan time could facilitate the routine implementation of post-LuPSMA SPECT/CT in clinical practice.

Supplementary Information

The online version contains supplementary material available at 10.1186/s40658-026-00864-6.

Introduction

Prostate cancer is the second most common cancer in men and ranks fifth in cancer-related mortality worldwide [1]. While localized prostate cancer generally has a favorable prognosis [2], outcomes for patients with metastatic castration-resistant prostate cancer (mCRPC) remain poor [3]. [¹⁷⁷Lu]Lu-PSMA-617 (LuPSMA) is a radiopharmaceutical therapy (RPT) for mCRPC that targets overexpressed prostate-specific membrane antigen (PSMA) on prostate cancer cells and delivers short-range beta-particle radiation (~ 1 mm in tissue), effectively damaging tumors with minimal impact on surrounding tissue [4]. In addition to beta-particles, ¹⁷⁷Lu emits gamma rays at 113 keV (6.2%) and 208 keV (10.4%) [5], allowing imaging of LuPSMA biodistribution with single-photon emission computed tomography (SPECT) combined with x-ray computed tomography (CT).

Whole-body post-therapy SPECT/CT imaging plays an important role for verifying LuPSMA uptake in tumors and organs relative to pre-treatment PSMA positron emission tomography(PET)/CT findings and in assessing treatment response. A recent retrospective study reported that LuPSMA SPECT/CT changed clinical management for nearly half of the patients imaged [6]. Furthermore, SPECT/CT-based dosimetry allows personalized treatment planning by quantifying absorbed dose in tumors and critical organs [7]. However, these multi-bed SPECT/CT acquisitions, typically lasting 30–45 min on conventional dual-head scanners, are resource-intensive and can be challenging for patients, who must remain still for prolonged periods. Multiple scans throughout treatment present a substantial burden for patients and clinical facilities. The recent FDA approval of LuPSMA before taxane chemotherapy is expected to increase the demand for post-LuPSMA imaging. Reduced scan times could facilitate more efficient and practical implementation in routine clinical practice.

While faster scans can be performed with the cadmium zinc telluride (CZT)-based SPECT/CT scanners equipped with 360° detector modules surrounding the patient [8–10], their limited availability and higher cost restrict widespread adoption. Therefore, achieving similar scan time reductions on conventional NaI(Tl)-based scanners is essential for broader clinical impact. Imaging on conventional scanners is typically acquired using a single photopeak energy window centered at 208 keV [11], disregarding the 113 keV emissions. However, despite its lower emission abundance, the 113 keV emissions can contribute nearly half of the total detected counts due to higher detection efficiency in standard 1-cm-thick NaI(Tl) crystals [12]. Thus, incorporating a dual-energy window acquisition could improve count statistics, supporting shorter scan durations.

The main challenge of imaging with the 113 keV window is downscatter from 208 keV photons, which may undergo single or multiple Compton scatters in the patient, collimator, or crystal (i.e., iodine escape) or penetrate the collimator and crystal without interaction and subsequently backscatter at ~ 115 keV [13], registering in the 113 keV window despite originating from 208 keV photons. Another issue is the contamination of 113 keV scatter window from characteristic x-rays (72–88 keV) from a lead-alloy collimator [13]. While standard energy-window-based scatter correction (EWS) is often inadequate for addressing these issues, advanced methods such as Monte Carlo-based scatter correction (MCS) can provide more accurate compensation, thereby improving image quality and quantitative accuracy. Studies using NEMA, anthropomorphic, and Jaszczack phantoms filled with ¹⁷⁷Lu [13–15], as well as realistic [¹⁷⁷Lu]Lu-DOTATATE SPECT/CT Monte Carlo simulations [12], have recommended using advanced scatter correction methods such as Effective Source Scatter Estimation (ESSE) [16, 17] and Analytical Photon Distribution–Interpolated (APDI) [18] for accurate quantification when employing 113 keV window, where EWS was not accurate.

Despite findings from phantom studies, clinical evaluation of the 113 keV photopeak or dual-energy acquisition on conventional scanners remains limited. This retrospective study evaluates the feasibility of reducing scan time in post-LuPSMA SPECT/CT imaging by employing a dual-energy approach with MCS across a range of scan durations.

Materials and methods

Study design and patients

This retrospective study was approved by the institutional review board with a waiver of informed consent (Dana-Farber Cancer Institute Protocol No: 20–259). We analyzed 33 post-therapy LuPSMA SPECT/CT scans acquired from 29 patients (two patients had scans at treatment cycle 1 and 3) treated as part of standard clinical care between May and September 2024. The median patient age was 70 years (range: 55–85), with a median injected activity of 7.3 GBq (range: 3.8–7.6), a median post-injection imaging time of 20 h (range: 1 h–3 days), and a median body mass index (BMI) of 29.2 (range: 19.9–40.4).

SPECT/CT acquisition and EWS reconstruction protocol

Whole-body LuPSMA SPECT/CT studies from mid-thigh to vertex were acquired using a conventional dual-head Symbia Intevo SPECT/CT scanner (Siemens Healthineers, Erlangen, Germany) with dual-photopeak and corresponding scatter windows as listed in Table 1. 208 and 113 keV photopeak windows (208 PP and 113 PP) and 208 keV lower scatter window (208 LW) were set to 20% and 10% width, respectively, in accordance with MIRD Pamphlet No. 26 [12]. Narrower scatter windows were applied adjacent to the 113 keV photopeak: the 113 keV upper scatter window (113 UW) was set to 10% to better match the scatter distribution within the photopeak, which includes both downscatter originating from 208 keV and scatter from 113 keV photons. The 113 keV lower scatter window (113 LW) was set to 5% to minimize contamination from characteristic lead x-rays. Note, the scatter windows were adjusted only for EWS, as MCS relies solely on the photopeaks.

Table 1.

¹⁷⁷Lu acquisition energy windows

Windows	Window width(%)	Energy window (keV)
208 Photopeak (208 PP)	20	187.2–228.8
208 Lower Scatter (208 LW)	10	166.4–187.2
113 Upper Scatter (113 UW)	10	124.3–135.6
113 Photopeak (113 PP)	20	101.7–124.3
113 Lower Scatter (113 LW)	5	96.0 – 101.7

Each bed position was scanned for 12 min in continuous mode (120 projections total) using auto-contouring over 360° with medium-energy-low-penetration (MELP) collimators. The total scan time for the three-bed acquisition, which covered up to 1117 mm axially with ~2 mm overlaps between adjacent beds, was approximately 40 min, including 36 min for SPECT, 2 min for bed transitions (1 min per translation), and additional time for the x-ray topogram and CT. Projections were reconstructed using the scanner’s Flash3D iterative algorithm (OSEM: 10 iterations 8 subsets, with point spread function modeling, no pre/post image filtering) with low-dose CT-based attenuation correction (AC) and EWS. Projections for the 113 and 208 keV windows were first independently reconstructed using their respective attenuation maps, then averaged into a single reconstruction.

Synthetic images at reduced acquisition time

Due to the lack of list-mode capability on the Intevo scanner, projections simulating shorter acquisition times could not be generated from a single acquisition. Therefore, we adopted an alternative approach in which synthetic projections for 1.5, 3, and 6 min/bed were created by applying Poisson noise to the original 12 min/bed projection data. Using a Python script, Poisson noise was separately applied to the five energy windows for each dataset. Each pixel value in the projection images was scaled to a lower count, and then noise was added by sampling from a Poisson distribution with a mean equal to the new lower pixel value. The modified and original projection data were then reconstructed using dual-energy MCS Hermia HybridRecon^TM-Oncology 4.0.6 software (Hermes Medical Solutions, Stockholm, Sweden) with the same parameters as in EWS, except that 10 iterations and 10 subsets were used in this case. In this MCS approach, scattered photons, including the downscatter are modelled for each projection view [19]. Independent reconstructions with MCS + AC from two energy windows were averaged into a single reconstruction as described above. As a substudy, 3 min/bed EWS reconstructions were also obtained for three patients to compare with the corresponding 3 min/bed MCS reconstructions. The 12 min/bed dataset reconstructed with MCS, served as the reference standard. This choice was supported by prior phantom studies demonstrating superior performance of advanced scatter correction methods over EWS [13–15, 18] and our initial visual evaluation of patient reconstructions.

Segmentation and data preparation

Organ and tumor segmentations were performed using the MIM SurePlan™ MRT version 7.3.3 (MIM Software Inc., Beachwood, Ohio). CT-based organ contours for bone, lungs, liver, kidney, and spleen were generated using MIM Contour ProtégéAI, an AI-based automatic segmentation tool, and transferred to the registered SPECT reconstructions. Tumor segmentations were derived directly from the SPECT images using the MIM’s gradient-based PET edge tool. All visible lesions with volumes ≥ 0.5 ml and demonstrating sufficient uptake on SPECT images were included.

Images reconstructed with MCS at 1.5, 3, and 6 min/bed scan durations were linearly scaled by factors of 8, 4, and 2, respectively, to match the reference count level at 12 min/bed. Given the inherent differences between EWS (Siemens Intevo) and MCS (Hermes) reconstructions, patient-specific scale factors were determined from whole-body mean values and applied to the EWS reconstructions. Further details in data preparation are presented in the supplemental file.

Image quality assessment

Image quality analyses were performed on SPECT reconstruction data based on bias, coefficient of variation (CV), contrast ratio (CR), and signal-to-noise ratio (SNR) metrics.

i) Bias in mean and maximum values was calculated for both organs and tumors by comparing each reconstruction to the reference dataset (MCS at 12 min/bed) as follows:

where test recon refers to the MCS at 1.5, 3, and 6 min/bed and EWS at 12 min/bed reconstructions. Bias was averaged across all organ and tumor contours in each dataset, using the absolute difference to prevent positive and negative values from canceling each other out.

ii) Coefficient of Variation (CV), defined as the standard deviation (SD) divided by the mean, was calculated in a 3-cm diameter spherical volume-of-interest (VOI) placed in the liver:

iii) Contrast Ratio (CR) of tumor to liver background, defined as the difference of the means in tumor and background in relation to mean of background (bkg), was calculated for each tumor where the same VOI used for CV served as background:

iv) Signal-to-noise ratio (SNR), defined as the difference of the means in tumor and background in relation to SD of the same background:

CR and SNR were calculated for each tumor individually, and their mean values were reported for each reconstruction condition. Statistical analyses comparing image quality metrics across different reconstruction methods were performed using the Kruskal–Wallis test in GraphPad Prism version 10.4.1.

Results

Counts in patient projections

Figure 1 displays anterior projection images from five energy windows for a representative patient scan acquired at 12 s/view (equivalent to 12 min/bed), along with corresponding total counts (summed over all projections) reported in millions (M). Although 113 keV photopeak yields higher total counts than the 208 keV photopeak, its image quality is poorer due to the downscatter. The scatter windows (208 LW, 113 UW, and 113 LW) show similar count levels despite different energy window widths.

Fig. 1.

Anterior projection images of the five energy windows from a representative 12 s/view (or 12 min/bed) patient acquisition, along with their corresponding total counts in millions (M). PP: photopeak window, LW: lower energy window, and UW: upper energy window

Figure 2 presents box plots of total projection counts across all 33 studies for each energy window and for separate acquisition beds starting from mid-thigh (bed 1) to vertex (bed 3), as well as total counts from all three beds. Variations in counts across the patient cohort are primarily due to the differences in post-injection imaging time (delta time) and, in some cases, lower administered doses. The count ratios between the two photopeak windows (113 PP / 208 PP) showed a mildly decreasing trend from thigh to vertex and remained relatively consistent among patients, with mean ± SD of 1.57 ± 0.08 for 1st bed (thigh–pelvis), 1.51 ± 0.04 for 2nd bed (pelvis–chest), and 1.46 ± 0.10 for 3rd bed (chest–head). This trend is consistent with decreased tissue attenuation from lower to upper body regions, corresponding to reduced amount of downscatter from the patient’s body. A similar trend was observed in patients with high BMI (BMI > 35), with higher ratios (1.64 ± 0.06, 1.57 ± 0.04, and 1.53 ± 0.03, respectively), again indicating the relationship between tissue attenuation and the extent of downscatter. Supporting this observation, the ratio of the 113 keV upper scatter window to its photopeak (113 UW / 113 PP), used as a surrogate for downscatter, also showed a similar decreasing pattern (0.23 ± 0.01, 0.22 ± 0.01, and 0.21 ± 0.02, respectively). Similar trends were observed in patients with BMI > 35, with higher ratios (0.24 ± 0.01, 0.23 ± 0.01, and 0.22 ± 0.01, respectively), suggesting that downscatter contribution is not strongly dependent on patient size. All differences across bed positions were statistically significant (p < 0.0001). The raw projection count data for the individual beds along with patient BMI, administered activity, and delta imaging time are provided in the supplementary data.

Fig. 2.

The box plots display the distribution of total projection counts (mean and standard deviation) in different energy windows across all the studies, for each bed and for all beds. PP: photopeak window, LW: lower energy window, and UW: upper energy window

Image quality evaluation at different scan times

Figure 3 shows sample images (maximum intensity projection: MIP, coronal, and axial) from a representative patient (injected activity:7.2 GBq, imaging: 20 h post injection, and BMI: 28.9) reconstructed using MCS at scan durations ranging from 1.5 to 12 min/bed (reference: 12 min/bed) and using EWS at 12 min/bed. All images are displayed with the same linear intensity scale (0–2200). For clarity, only tumor segmentations are shown on the MIP images. As expected, visual image noise decreases with longer scan durations. Although both the reference and EWS reconstructions used full-count data and identical energy windows, the reference MIP (MCS 12 min/bed) appears less noisy and shows slightly better contrast.

Fig. 3.

Sample MIP, coronal, and axial images (injected activity: 7.2 GBq, delta time: 20 h, and BMI: 28.9) showing Monte Carlo-based scatter correction (MCS) at 1.5–12 min/bed (reference: 12 min/bed) and energy-window-based scatter correction (EWS) with 12 min/bed. Tumor segmentations are shown in MIP images. Tumors indicated on the coronal and axial reference images can be clearly seen in all the reconstructions. In the 1.5 min/bed reconstruction, hot spot in the liver (indicated by red dashed circle) may be falsely interpreted as tumor (false positive)

Across all 33 scans analyzed, a total of 327 lesions were evaluated. Figure 4 summarizes bias in mean and maximum measurements in organs and tumors for MCS and EWS. In panel A, each data point represents the % bias for a patient study averaged over organs and tumors (asterisks indicate statistical significance). Panel B shows violin plots illustrating the distribution of % bias in mean and maximum across all organs and tumors. The average bias in mean values for the MCS reconstructions at 1.5–6 min/bed was less than 6% for organs (median ~ 2%) and at 3–6 min/bed less than 10% for tumors (median ~ 5%) relative to the reference. These measurements were higher for EWS at 12 min/bed with median ~ 3% for organs (p = 0.003–0.01) and ~ 6% for tumors. The average bias in organ maximum values for MCS at 1.5–3 min/bed were higher than that of EWS (p < 0.0004), whereas the differences between MCS and EWS for the bias in tumor maximum value were not significant. For EWS at 12 min/bed, lower bias was observed in organ and tumor maximum values compared with low-count MSC, whereas higher bias was observed for corresponding mean values. This suggests that noise-related effects have a stronger influence on maximum values, while mean values are more strongly affected by overall image contrast and reconstruction characteristics.

Fig. 4.

A Box plots show mean and SD for the average bias in mean and max measurements for organs and tumors with the Monte Carlo-based scatter correction (MCS) and energy-window based scatter correction (EWS). Each data point corresponds to a patient’s bias averaged over absolute mean and max measurements (asterisks indicate statistical significance: **p < 0.01 and ****p < 0.0001). B Violin plots show the distribution of bias in mean and max for all the organs and tumors. In this case both positive and negative bias are shown without using the absolute function

Similar bias trends were observed across individual organs (bone, lungs, liver, kidneys, and spleen). Among these, the spleen exhibited the highest mean % bias for all reconstructions (Supplemental Fig. 4). Using the median tumor volume (5 ml) as a cutoff, bias was higher for smaller tumors (< 5 mL) compared with larger tumors (> 5 ml) (Supplemental Fig. 5), consistent with reduced count statistics in smaller lesions. For MSC reconstructions, the difference in bias between small and large tumors decreased with increasing scan time, as quantified by the ratio of bias between small and large tumors: ratio of 1.5 at 1.5 min/bed, 1.3 at 3 min/bed, and 1.2 at 6 min/bed. This trend is consistent with smaller tumors being more susceptible to count deficiency and showing greater improvement with increased count statistics. In contrast, for EWS reconstructions, the bias ratio (1.3 at 12 min/bed) was similar to that observed for MSC at 3 min/bed despite higher count statistics. Overall, this suggests improved performance of MSC relative to EWS for smaller tumors.

Figure 5 presents CV, mean(CR), and mean(SNR) measurements. For clarity, extreme outliers (very low CV and very high CR and SNR) were excluded from the figure. As an indicator of image noise, lower CV values are preferred. The reference CV was lower than all other reconstructions, with statistically significant differences only compared to the 1.5–3 min/bed reconstructions. For CR and SNR, higher values are preferred. Mean(CR) did not differ significantly across reconstruction methods, whereas mean(SNR) was the highest in the reference, with significant differences only compared to the 1.5–3 min/bed reconstructions.

Fig. 5.

Coefficient of variation (CV) was calculated in a 3 cm diameter spherical VOI in liver. Mean contrast ratio (CR) and mean signal to noise ratio (SNR) were calculated from tumors (signal) and from the same liver VOI (background). Each data point corresponds to data from one patient. Some of the outliers were not included in the plots for display purposes (asterisks indicate statistical significance: *p < 0.05, **p < 0.01,***p < 0.001 and ****p < 0.0001)

At high count levels, the advantages of MCS over EWS were less apparent. Figure 6 presents MIP images from three representative patients with injected activities of 3.8, 6.2, and 7.2 GBq for visual comparison of EWS and MCS reconstructions at 3 and 12 min/bed acquisition times. For the 3 min/bed, MCS images exhibit visibly improved contrast and reduced noise compared to EWS (EWS and MCS at 3 min/bed used the same noisy projections). In contrast, the differences between EWS and MCS are more subtle at 12 min/bed where image quality is generally higher for both methods.

Fig. 6.

Sample MIP patient images comparing EWS and MCS for 3 min/bed and for 12 min/bed. Contrast and noise in MCS are visually better than that of EWS for 3 min/bed. The differences between EWS and MCS are more subtle for the 12 min/bed. The displays were auto-normalized for each MIP individually

Discussion

Post-RPT SPECT/CT has proven valuable in the management of patients undergoing LuPSMA treatment [21–23], motivating the need for more streamlined and practical imaging protocols. With growing RPT use, SPECT is increasingly utilized for response assessment, especially when PET access is limited. This underscores the need for simpler, shorter, and more patient friendly SPECT imaging. However, standardized guidelines for the post-LuPSMA SPECT/CT imaging protocols are lacking, with practices varying widely across institutions, ranging from 5 to ~ 40 min/bed for 2–3 bed positions [7, 23–30]. Dosimetry is not yet routinely implemented. For practical dosimetry, Madsen et al. [31] and Hänscheid et al. [32] introduced the theoretical foundation for single-time-point dosimetry, which has since been successfully implemented by several groups for ¹⁷⁷Lu dosimetry [33–36]. In line with these efforts, we implemented protocol improvements aimed at reducing scan duration while maintaining quantitative accuracy and reasonable image quality for practical post-LuPSMA SPECT/CT and dosimetry. The continuous acquisition mode reduced the scan time by 3 min per bed position (a total of 9 min reduction for a 3-bed scan), which is particularly relevant for shorter scans. In this mode, detectors acquire data continuously throughout both rotational and radial motion providing approximately 3 min of additional counting time compared to a 60-step standard step-and-shoot acquisition, where data collection occurs only when the detectors are stationary. This counting time gain is proportional to the number of rotational steps (views) and independent of the scan duration per view. The dual-energy acquisition increased detected counts and MCS provided more accurate scatter compensation, especially under downscatter and low-count conditions where EWS performs poorly. The current version of the MCS method did not include collimator scatter modeling for ¹⁷⁷Lu, which could further improve the scatter correction. Advanced noise reduction strategies, such as those leveraging deep learning approaches [23, 24, 37, 38], could further improve noise in low-count situations.

A series of publications from the SNMMI Lu-177 Dosimetry Challenge 2021 highlighted substantial variability in dosimetry estimates across participants [39–42]. Given the many sources of uncertainty in dosimetry, scan duration may not be as critical a contributor to overall error. Mean dose, averaged over dose values within tumor or organ, is typically reported for dose estimation and is closely related to mean uptake (or voxel-regions) [43], which tends to remain robust even in noisy reconstructions, as demonstrated in our study by bias measurements showing reasonable accuracy down to 1.5–3 min/bed scan durations. While maximum values were more sensitive to count levels and exhibited larger biases, they remained within an acceptable range (e.g., median bias: ~10% for tumors and organs at 3 min/bed) and their quantitative accuracy is less relevant for dosimetry.

Although high-quality images are always more desirable, in practice, scan duration and image quality should be balanced to meet diagnostic needs, while also considering scanner availability, patient comfort, and the increased risk of motion artifacts with longer acquisitions. For image quality assessment, we obtained CV, CR, and SNR metrics across scan durations (1.5–12 min/bed) for MCS and EWS at 12 min/bed. In terms of these metrics, MCS at 6 min/bed was not significantly different from either full-duration (12 min/bed) reconstruction. Visual assessment supported these findings, showing reasonable image quality for MCS at 3–6 min/bed (Fig. 3), though lesion detection at 1.5 min/bed was more challenging, potentially leading to false positives or missed lesions. Notably, the visual image quality was substantially better for MCS than EWS at 3 min/bed (Fig. 6), indicating the performance improvement with the MCS method over EWS at low count levels. Additionally, MCS at 3 and 12 min/bed showed comparable overall radiotracer distribution, with most lesions still identifiable, although with reduced certainty due to the increased noise. As discussed below, real-world image quality of short acquisitions would be slightly better than our simulated results. For shorter scans, image filtering and lower number of iterations can also be considered to suppress the noise, which would improve CV and SNR metrics as well.

In this study, we also provided a detailed projection count distribution across patients for the three beds separately and five different energy windows. We expect these data will be valuable for investigators in evaluating or optimizing their acquisition and reconstruction protocols. With our current energy window settings, we achieved similar count levels for all scatter windows despite their varying widths, ensuring sufficient counts for reliable scatter estimation. Downscatter accounts for a substantial portion of the detected counts in the 113 keV photopeak window, which was the main reason for using the MCS approach in this work. To further evaluate this effect, we analyzed the 113 PP / 208 PP and 113 UW / 113 PP count ratios, across the bed positions and patient BMIs. These ratios were only ~ 5% higher in patients with BMI > 35 compared to those with BMI < 25, indicating only a mild relationship between the tissue attenuation and extend of downscatter. This modest trend likely reflects two main competing effects with increasing tissue thickness: increased attenuation of primary 113 keV photons and increased downscatter from higher-energy photons into the 113 keV window, which partially offset each other.

While MSC and EWS are implemented using reconstruction frameworks from different vendors, reconstruction parameters were kept identical wherever possible, except for the number of subsets (MSC: 10 iterations × 10 subsets; EWS: 10 iterations × 8 subsets). In this study, the effective number of iterations was intentionally not adjusted as a function of count level. As a result, while the difference in subsets may have a modest influence on quantitative metrics, its impact is expected to be secondary to the effects of count reduction. Supporting this, a substudy performed using MSC that compared two effective iteration settings (100 vs. 80) across 15 lesions demonstrated a mean absolute difference of < 2%.

One limitation of this study was the lack of list-mode data. To simulate shorter scan durations, we applied Poisson noise to the clinical projections, although ideally this should have been applied to noise-free projections. As a result, the simulated low-count projections became noisier by a factor of 1.225 [44] than actual clinical projections at equivalent count levels. In fact, our phantom study with uniform activity (supplementary material) indicated that the reconstruction was noisier for the synthetic data (~ 20% higher SD and CV) compared to the real data. Therefore, we anticipate that real-world performance with shorter scans would be slightly less noisy than our simulated results. We also considered generating low-count datasets by discarding a subset of the projection frames, rather than altering the original count distribution with added noise. For example, to simulate 1.5, 3, and 6 min/bed scans, we would retain only 15, 30, and 60 projections, respectively, from the full 120-projection acquisition. Reconstructions using this method were noticeably less noisy than those generated using Poisson noise, supporting the expectation that real-world low-count projections will perform better than our simulations. However, several issues led us to prefer the Poisson-based approach. While discarding projections preserves the statistical properties of the acquired data, it also removes temporal variation, such as patient motion (e.g., respiratory and body motion), that is present in the full dataset. This alters the overall distribution of counts and can introduce mismatches with the full-count reconstruction. In fact, we observed substantial non-rigid misalignment in tumor locations in several cases using this approach. These temporal inconsistencies also impact attenuation and scatter corrections, which would further introduce discrepancies with the full-count reconstruction. It is worth noting that similar issues could arise even with list-mode data if the reduced dataset is not sampled randomly. Additionally, reduced angular sampling by removing projections may introduce artifacts unrelated to scan duration. There are also some implementation challenges with this approach. Reduced number of projections requires modification of numerous DICOM tags to ensure the reconstruction software properly interprets the modified data (a Python script provided in the supplementary material). Considering all these factors, the Poisson noise approach proved to be far simpler, required minimal modification, and was ultimately more suitable for the objective of this study.

Conclusion

This study demonstrates the feasibility of reducing post-LuPSMA SPECT/CT scan durations on conventional NaI(Tl)-based scanners by employing dual-energy window acquisition and Monte Carlo-based scatter correction. Quantitative metrics and visual assessments indicate that scan times on conventional NaI(Tl)-based scanners as short as 3–6 min/bed may be adequate for clinical tasks such as lesion detection, therapy monitoring, and dosimetry. Further improvements in image quality and reduction in scan duration may be achieved through optimization of reconstruction parameters, refinement of Monte Carlo-based scatter correction method, and application of deep learning-based noise-reduction techniques.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

Arda Könik (PhD) designed the study, collected the data, and performed the analyses. Benjamin Auer (PhD) contributed to the study design and assisted with data generation. Shir Hazut Krauthammer (MD) provided clinical support including visual evaluation and assisted with segmentation, Praful Ravi (MD) provided clinical support, Thomas S.C. Ng (MD, PhD) contributed to the study design, assisted with analyses, and critically reviewed the manuscript, Heather A. Jacene (MD) gave clinical support including visual evaluation. All authors critically reviewed and approved the final manuscript.

Funding

No funding was received for this study.

Data availability

The datasets generated and analyzed for this study are available from the corresponding author and some of the datasets were shared as supplemental data in the submission.

Declarations

Ethics approval and consent to participate

This retrospective study was approved by the institutional review board with a waiver of informed consent (Dana-Farber Cancer Institute Protocol No: 20–259). Patient consent was waived due to the retrospective nature of the study and use of anonymized data.

Consent for publication

Not applicable.

Competing interests

The authors declare that they have no competing interests.