Dual Energy Imaging for Optimisation of P/V Ratios in VP SPECT


 PurposeVentilation-perfusion single photon emission computed tomography (VP SPECT) plays an important role in pulmonary embolism diagnosis. Rapid results may be obtained using same-day ventilation followed by perfusion imaging, but generally requires careful attention to achieving an optimal count rate ratio (P/V ratio) of ~3:1. This study investigated whether the ratio of counts simultaneously acquired in adjacent primary and Compton-scatter energy windows (E_ratio) on V SPECT was predictive of final normalised perfusion count rate (PCRnorm) on P SPECT using [99mTc]Tc-macroaggregated albumin (MAA), allowing optimisation of P/V ratio. MethodsSame-day VP SPECT studies acquired using standard protocols in adult patients during a 2-year period were assessed. Studies were included provided they were acquired with correct imaging parameters, and injection site imaging and laboratory records were available for quality control and normalised count rate corrections. Extraction of DICOM information, and linear regression were performed using custom Python and R scripts. A predictive tool was developed in Microsoft Excel.ResultsOf 643 studies performed, the scans of 343 participants (median age 30.4 years, 318 female) met inclusion criteria. A single outlier with high influence was excluded after an obvious cause was identified. Final analysis of the remaining 342 scans yielded a significant regression equation (F(1,340) = 1057.3, p <.000), with an adjusted R2 of .756 and MSE of 0.001089. A prediction tool designed for routine clinic use was developed for predicting final P/V ratio.ConclusionThe ratio of simultaneously acquired counts in adjacent energy windows on V SPECT is linearly related to perfusion count rate after administration of a known activity of [99mTc]Tc- MAA. A predictive tool based on this work may assist in optimising the dose and timing of [99mTc]Tc-MAA administration in same-day studies to the benefit of patients and workflows.


Introduction
The perfusion component of a ventilation-perfusion single photon emission computed tomography (VP SPECT) study is typically undertaken on the same day as, and frequently immediately following the ventilation component. When using technetium-99m ( 99m Tc)-based radiopharmaceuticals in a same-day imaging strategy (generally recommended, especially to obtain rapid diagnosis of pulmonary embolism) [1], residual ventilation counts contribute to the acquired perfusion images and may obscure perfusion defects leading to misdiagnosis. While software can perform correction for residual ventilation counts in the nal perfusion image, slight differences in radiopharmaceutical distributions, and imperfect coregistration can lead to signi cant artefacts and it is therefore recommended that the perfusion count rate (PCR -corrected for residual ventilation counts) exceeds the ventilation count rate (VCR) by a factor of at least 3 [2][3][4]. This is equivalent, when a perfusion SPECT is performed immediately after the ventilation SPECT, of the uncorrected count rate exceeding the VCR by a minimum factor of 4. At the same time, higher ratios may also be considered undesirable since these would require high doses of the The targeting of an optimal perfusion/ventilation count rate ratio (P/V ratio) is complicated principally by the fact that it is often di cult to quantify the amount of ventilation activity that has been administered in any individual patient. In patients with a large amount of attenuating soft tissue, higher amounts of administered activity may be required to obtain a su ciently high VCR to guarantee a ventilation SPECT of su cient quality. Post hoc calculation of administered ventilation activity using knowledge of the imaging system sensitivity alone is not possible due to the frequently signi cant amount of attenuation by thoracic tissue. While absolute quanti cation of administered ventilation activity is possible using quantitative SPECT, this remains experimental and requires specialised software and a SPECT/CT system [5]. As a result, the dose and timing of [ 99m Tc]Tc-MAA administration after the ventilation SPECT is frequently suboptimal, resulting in PCRs that are either too low to achieve an adequate P/V ratio, or the administration of activities exceeding those recommended or required, and thus not meeting the principle of "as low as reasonably achievable" (ALARA).
Using a variety of techniques, it is possible to account for photon attenuation using information from a second, simultaneously acquired Compton scatter window [6]. A linear relationship between the number of photons detected in scatter windows and voxel attenuation coe cients has previously been described [7].
This study aimed to investigate whether a simple ratio of counts acquired in two adjacent energy windows on the ventilation SPECT (E_ratio) is predictive of the nal perfusion count rate normalised for [ 99m Tc]Tc-MAA activity (PCR norm ) on the perfusion SPECT; and if so, whether such a relationship can be used to develop a simple tool to adjust either the activity (dose) of [ 99m Tc]Tc-MAA administered or the timing (or both) of the perfusion SPECT, such that an optimal P/V ratio may be achieved.

Participant selection
All patients referred to the Tygerberg Hospital Division of Nuclear Medicine for VP SPECT studies between July 2017 and July 2019 were eligible for inclusion. Only patients with same-day studies (ventilation followed by perfusion) and who were older than 18 years were included in this retrospective analysis. Patients in whom no [ 99m Tc]Tc-MAA injection site imaging was performed; or for whom the [ 99m Tc]Tc-MAA dispensing time was not recorded (to correct [ 99m Tc]Tc-MAA dose for residual counts at administration site, and for decay) were excluded, as were the scans of patients in whom the imaging acquisition differed from the divisional protocol; or in cases in which a repeat SPECT acquisition was performed.

Imaging procedure
Ventilation was performed with the patient lying supine, using a [ 99m Tc]Tc-Technegas generator (Vita Medical Ltd, Lucas Heights, New South Wales, Australia). After each breath, tubing of the ventilation system was set aside and a Geiger-Müller (GM) detector (Radiation Alert® -Inspector, SE International Inc, Summertown, TN, United States) positioned on the sternum. Technicians were instructed to target an approximate stabilised probe count rate of ~2500 counts per minute (cpm). Immediately after ventilation, SPECT was performed on a dual-head Siemens Symbia SPECT camera (Siemens Healthineers, Hoffman Estates, IL, USA). Acquisition protocols were chosen to match most closely those recommended by Palmer et al [4] - Table 1. All SPECT scans were acquired with a primary energy window centred on a peak of 140 keV (±7.5%); and a symmetrical down-scatter window, centred on a peak of 119 keV, of exactly the same width immediately beneath this. The latter window has been routinely included in VP SPECT acquisitions in the Tygerberg Division of Nuclear Medicine since 2015, at which time image reconstruction software (Hermes Gold 3; Nuclear Diagnostics, Stockholm Sweden) used this data to construct a synthetic attenuation map.
The perfusion study was acquired either immediately after the ventilation study or after a delay to allow for some decay of ventilation counts using ~120 MBq of [ 99m Tc]Tc-MAA (Pulmocis®, Cisbio). All dispensed [ 99m Tc]Tc-MAA doses had a radiochemical purity of ≥95%. Static acquisitions (64 x 64 matrix, 60 seconds, 129.5 -150.5 keV), using the same LEAP collimator as used for SPECT, were made of the [ 99m Tc]Tc-MAA injection site to exclude major misinjection and to quantify residual activity in the injection portal. Residual activity in the syringe was not measured.

Data extracted from DICOM les and paper records
From the radiopharmacy and patient scan records, the time [ 99m Tc]Tc-MAA was dispensed and the activity at dispense time were recorded.
All raw tomographic VP SPECT image data, as well as corresponding injection site images for the period in question were downloaded from the divisional PACS in DICOM format.
A custom script written in Python version 3.8.2 [8] using the pydicom package version 1.4.2 [9] was used to extract information from the DICOM image and header tags, both to assure compliance with divisional acquisition protocols, and to obtain the necessary image measures for analysis purposes.
Information to ensure consistency of the acquisition protocols included width and centring of the energy windows; scan arc; number of frames; frame duration; and camera and collimator identi ers. Scan order (ventilation followed by perfusion) was also validated by checking scan dates and times.
Measures extracted for analysis purposes included patient name, surname, and folder number (subsequently deidenti ed); patient gender and date of birth (for demographic purposes); scan dates and times (for decay correction); and total image counts for both energy windows (for tomographic data) or in the primary energy window for the injection site images.
Derived image measures and how they were calculated are summarised in Table 2.

Statistical analysis
A linear regression model of E_ratio and PCR norm was built and tested in RStudio version 1.3.1093 [10] using R version 4.0.2 [11]. Diagnostic plots were generated using native and ggfortify [12] packages. A threshold of p=0.05 was considered statistically signi cant. Calculation of 95% prediction intervals for graphing purposes was performed using the native predict.lm function.

Prediction tool
Using the results of the linear regression, and the mean squared error (MSE), a tool to predict nal P/V ratio was developed in Microsoft® Excel® for Microsoft 365 MSO (16.0.13127.20620) 64-bit.

Selection of participant scans
A total of 643 studies (ventilation SPECT followed by same-day perfusion SPECT) were identi ed in the study period. Of these, 22 were of patients under the age of 18 years and were excluded. An additional 199 studies were excluded due to missing injection site image; 1 signi cant misinjection (83 MBq); 69 due to missing radiopharmacy records ([ 99m Tc]Tc-MAA dispense time not recorded); 4 due to incorrect dispense time records (recorded time later than scan time); and 1 due to deviation from the divisional protocol (incorrect collimator used). Four participants had repeat studies (10.1 -58.6 weeks later), and for these participants, only the rst study was included in the analysis. An initial analysis was performed on 343 imaging studies. One imaging study was retrospectively excluded after it became apparent that it was an outlier with high in uence and an obvious cause was identi ed (details below). The sample size for the nal analysis was thus of 342 imaging studies.

Demographics and summary statistics
Of the 342 imaging studies included in the nal analysis, 317 were performed in female patients. The median patient age was 30.1 years (IQR: 25 -37.2 years). Mean A sys was 114.8 MBq (SD: 8 MBq). Mean E_ratio was 1.83 (SD: 0.21), and mean PCR norm was 0.313 (SD 0.067). Mean P/V ratio was 4.01 (SD 1.09). A breakdown of the nal P/V ratio values in the study sample appears in Table 3.

Linear regression
A simple linear regression was calculated to predict PCR norm , given E_ratio. The initial scatter plot and diagnostic charts (Figure 1) identi ed an obvious outlier (datapoint 223: x= 1.89, y=0.62). Further investigation revealed that the ventilation SPECT in this case had a very high amount of activity outside of the patient which very likely in uenced the E_ratio value ( Figure 2). Given that this observation was both an outlier and highly in uential, and because an obvious cause was identi ed on review of the raw data, this observation was retrospectively excluded.
After exclusion of the outlier, a signi cant regression equation was found (F(1,340) = 1057.3, p <.000), with an adjusted R 2 of 0.756 and MSE of 0.001089. After adjusting for any speci c camera-collimator sensitivity factor, the relationship to yield a re-normalised (adjusted) perfusion count rate was represented by the following equation: where PCR represents the expected perfusion count rate (in cps), for any administered activity (A) of [ 99m Tc]Tc-MAA (in MBq), re-normalised (adjusted) to account for the user-speci c sensitivity of the camera-collimator combination , and predicted by E_ratio.
The relevant diagnostic plots for the nal regression appear in Figure 3. The nal model, including con dence and prediction intervals is plotted in Figure 4.

Prediction tool
The output of the linear regression enables calculation of the prediction interval around y, for any future value of x using the equation: where is the predicted tted value of for a particular value of x; represents the two-tailed inverse of the Student's t-distribution with signi cance level for number of observations in the regression dataset; MSE is the mean squared error; X h is is the value of a hypothetical future observation of X; is the mean value of X in the regression dataset; and represents the respective values of in the regression dataset [13].
Using the above formula, a practical tool to predict the nal PVR (with 90% and 95% prediction intervals) was developed in Excel (Supplementary Materials 1). A video demonstrating practical implementation of the tool is included in Supplementary Materials 2.

Discussion
This study found a linear relationship between the ratio of ventilation counts in the primary energy window and an adjacent Compton scatter window and the nal perfusion count rate after the administration of a given dose of [ 99m Tc]Tc-MAA . Results of the analysis were used to develop a practical, Excel-based tool to predict the adequacy of the P/V ratio in same-day VP SPECT studies, which nuclear medicine practitioners can use to optimise the trade-offs that must frequently be made in these studies.
In a same-day VP SPECT study, the ventilation component is followed (ideally immediately after) by the perfusion component. Different strategies are used to increase the likelihood of obtaining ventilation images that are of su ciently high diagnostic quality, while still meeting a nal target P/V ratio of ≥3, and adhering to recommended [ 99m Tc]Tc-MAA doses. Most divisions target a speci c count rate (e.g. 2000 -2500 cpm) [14,15] measured using a GM detector after ventilation, and having achieved this, proceed to ventilation imaging. Depending on the count rate on the nal ventilation SPECT, centres either proceed immediately with injection of [ 99m Tc]Tc-MAA under the camera detector, usually checking for or titrating to a four-fold increase in counts, or delay the perfusion component of the study until such a time that ventilation count rate is judged to have su ciently decayed.
There are several problems with these strategies. Foremost, that the VCR is attenuated by variable amounts of soft tissue. This has the effect that a given activity (e.g. 40 MBq) of administered ventilation agent might result in a very high VCR (in a thin patient with minimal attenuating soft tissue), or a very low count rate (in a patient with more thoracic soft tissue), typically requiring the additional administration of ventilation agent. Very high count rates on the nal ventilation study may lead to unnecessary deferment of the perfusion study in the rst patient (in whom immediately proceeding with e.g. a 120 MBq dose of [ 99m Tc]Tc-MAA might in actual fact have been su cient), or lead to an inadequate P/V ratio in the second patient (who might require e.g. 80 MBq of administered ventilation agent to achieve a target count rate) when their perfusion study is undertaken immediately, even with the maximum recommended [ 99m Tc]Tc-MAA dose of 160 MBq [1]. Titrating the dose of [ 99m Tc]Tc-MAA administered to count rate under the camera is a useful compromise [16] but may require excessive amounts of activity (and particles) with attendant risks. It may be speculated that the di culties in optimising the trade-offs between dose, count adequacy, su cient P/V ratio, and reasonable patient turnaround times, are greatest in units caring for patients with wide variability in body habitus. Certainly, these trade-offs are frequently imperfect: a multicentre survey including the results of 286 VP SPECT studies in Germany reported inadequate P/V ratios in 25% of studies [16] which is similar to the ndings of the current study, in which 18% of studies did not meet the recommended P/V ratio of >3.
A P/V ratio that is too high is also problematic. While the optimal P/V ratio is uncertain, it seems unlikely that much additional diagnostic bene t is achieved with ratios above 5 (17 % of cases in the current analysis). While [ 99m Tc]Tc-MAA is generally accepted as a safe compound, it is not entirely innocuous, with reports of respiratory complications and death when high particle numbers have been administered to patients with underlying pathology of the pulmonary vasculature [17][18][19][20]. The risks posed by exposure to low doses of ionizing radiation are more controversial [21,22] but certainly it is reasonable to try and minimise the risks conferred by both particle number and radiation when reasonably achievable.
Most SPECT cameras can be easily con gured to simultaneously image two adjacent energy windows.
Once the ventilation SPECT is complete, it is a simple matter to perform a quick quality check of the raw data, and to use the perfusion adequacy tool to predict the nal P/V ratio. Moreover, this tool allows users to prioritise lower radiation dose or higher scan quality by modelling the impacts of different [ 99m Tc]Tc-MAA doses and/or delays before performing the perfusion SPECT on the P/V ratio.
This study suffered from several limitations. Firstly, many scans had to be excluded from the analysis due to missing injection site imaging or radiopharmacy records, which affected the nal sample size. Secondly, the sample was mostly composed of young (adult), pregnant women which means that caution should be exercised when applying the results to other patient demographics. Similarly, the linear relationship between E_ratio and PCR norm at extreme values of E_ratio was based on only a few observations, and the validity of the linear model at these extremes is less certain. Thirdly, con dence and prediction intervals may have been narrower had this study included additional corrections potentially affecting the accurate calculation of the amount of systemically administered [99mTc]Tc-MAA(A sys ): a) for residual activity in the syringe after administration, and b) for the interval between injection and the perfusion SPECT. The latter limitation is however not expected to be a major contributor to error, given the divisional policy of commencing the perfusion SPECT immediately post-injection. Fourthly, although radiochemical purity of [ 99m Tc]Tc-MAA was routinely above 95%, variation above this threshold was not accounted for, which may have contributed slightly to error in PCR norm . Finally, an inherent limitation in predicting PCRs is presented by variability in right-left shunting (including physiological shunting) which could not be accounted or corrected for [23].

Conclusion
The expected perfusion count rate for a given dose of [ 99m Tc]Tc-MAA is linearly related to the ratio of ventilation counts in two adjacent energy windows. This relationship allows for the application of a simple tool to predict and/or optimise the nal P/V ratio. Abbreviations: LEAP -Low-energy, all purpose; keV -kiloelectron volts; sec -seconds.   Figure 1 Diagnostic plots of initial sample. a Scatter plot of observations in which E_ratio represents the calculated ratio between acquired counts in the primary energy window divided by the down-scatter window on the ventilation study and PCRnorm represents the perfusion count rate per megabecquerel of administered [99mTc]Tc-MAA (cps/MBq), normalised to a camera-collimator sensitivity of 1 cps/MBq. b Chart of standardised residuals identi ed an obvious outlier (observation 223), which was also shown to be highly in uential on c Cooks' distance chart. Investigation of this observation identi ed an obvious cause (Figure 2), and on this basis the case was excluded and the regression repeated. Images generated in RStudio. Figure 3