Development and validation of a pulmonary embolism quantification method using single photon emission computed tomography ventilation and perfusion scintigraphy

Abstract

Background

Pulmonary embolism (PE) remains an important cause of cardiovascular death and morbidity worldwide. Conventionally, PE imaging is reported in a binary manner, with reports stating the presence or absence of a thrombus. No generally accepted graduated quantification systems are available to guide management or predict the development of acute or long-term complications. V/Q SPECT is a well-established method for diagnosing PE in acute and chronic settings. The use of volumetric perfusion and ventilation functional images could potentially calculate the volumetric PE burden. This project aimed to develop and optimise a volumetric V/Q SPECT PE quantification technique, utilising V/Q SPECT phantoms, 100 consecutive clinical V/Q SPECT studies, and advanced V/Q SPECT image analysis with semiautomated %PE quantification methods.

Results

Lung phantoms with PE defects of known volume demonstrated that the choice of acquisition, reconstruction parameters, and the volumetric cutoff threshold influences %PE results. The V/Q quotient image utilised a semiautomated technique with phantom-derived optimisation parameters, which was found to be the most accurate quantification method.

Conclusion

This study, combining phantom and clinical study analysis, developed a successful method for semiautomated V/Q SPECT %PE quantification, which has the potential utility of guiding patient management and outcome prediction.

Supplementary Information

The online version contains supplementary material available at https://doi.org/10.1186/s13550-026-01379-y.

Introduction

Pulmonary embolism (PE) is a treatable but potentially life-threatening condition identified as a common cause of morbidity and mortality worldwide. The clinical features of PE are often nonspecific [1, 2], and many PE-related deaths are only diagnosed at the post-mortem [3]. In the absence of reliable clinical symptoms and signs, imaging has become the cornerstone of PE diagnosis. Computed tomography pulmonary angiogram (CTPA) and Ventilation-Perfusion scintigraphy (V/Q) are the currently favoured diagnostic PE modalities.

In conventional clinical practice, diagnostic imaging for PE has been reported in a binary manner, stating the presence or the absence of a thromboembolus, supplemented with a general description of the size. However, in the recently published literature, there has been interest in developing methods for the accurate quantification of PE clot burden. This utilizes different imaging modalities to allow patient risk stratification, provide a more individualised approach to patient management and prognostication [4–6].

PE clot burden quantification can be achieved in two ways: by assessing the size and distribution of the thromboemboli within the arterial system (e.g., CTPA) or by assessing the effect of the thromboemboli on parenchymal perfusion (e.g., V/Q scintigraphy).

In V/Q scintigraphy, the size and number of the technetium-labelled microaggregated albumin particles used to demonstrate the parenchymal perfusion are just large enough to get wedged within the alveolar capillary bed to display an accurate map of pulmonary parenchymal perfusion. V/Q based PE quantification may be more pertinent than CTPA, as the clot burden estimation from CTPA angiographic images, which has dominated the literature [7–10] so far, is likely to underestimate the subsegmental burden due to limitations of linear resolution preventing the demonstration of clots within subsegmental arteries. V/Q scintigraphy, which shows parenchymal burden including subsegmental PE, is therefore likely to reflect the pathophysiological impact more truly.

To date, only a few studies have been published that have quantified V/Q scintigraphy-based PE burden [5, 11–14]. Even these have been mainly based on visual assessment scores, and V/Q SPECT has been shown to be better in delineating mismatched defects and more accurate in quantification compared to planar [12].

Compared to visual scoring, where the reporter is required to visually determine the size and number of perfusion defects compared to the ventilated lung volumes [12, 13], an objective, quantitative volumetric measurement would provide a more accurate, reproducible assessment of the perfusion defect.

This study aims to provide an objective assessment of the %PE parenchymal burden by building a semi-automated volumetric parenchymal %PE quantification method based on V/Q SPECT datasets shown to be accurate in comparison with lung phantoms containing a ‘phantom PEs’ of known volume.

Materials and methods

We used the following evaluations to establish the optimum acquisition and reconstruction parameters, method of quantification, definition of quantitative volumes of interest (VOI), and cutoff thresholds for accurate assessment and calculation of %PE, utilising two designs of V/Q SPECT lung phantoms. Then, the developed methods were applied to V/Q SPECT scintigraphy datasets to assess their accuracy in clinical cases.

1. Identification of optimum V/Q acquisition parameters using the Leeds Test Object V/Q lung phantom

This work was carried out during the COVID-19 pandemic, when it was not possible to carry out research scans, so previously acquired datasets were used to evaluate the optimum acquisition parameters. The Leeds Test Objects V/Q lung phantom was originally developed to allow optimisation of V/Q SPECT acquisition scanning parameters on different gamma camera and collimator combinations (Fig. 1) [15]. This phantom utilises a specially sourced sponge that allows free diffusion of ^81mKr after the sponge is immersed and compressed to squeeze out ^99mTc-microaggregated albumin (MAA), mimicking the lung tissue during V/Q scintigraphy. The square sponge contains 4 cylindrical defects (two large and two small) of known volume and is surrounded by perspex to resemble the scattering of body soft tissues. One large and one small cylindrical defect were left open to mimic mismatched defects of PE, and one large and one small cylindrical defect were blocked with sealed plastic jars to mimic matched defects.

Fig. 1.

VQ imaging phantom (anterior and anterior/oblique views) and positioning on the patient couch

The sponge was prepared by immersing it in 200 ml water containing 200 MBq ^99mTc pertechnetate, which was calculated to result in a similar concentration of activity as in clinical cases. Once the sponge was placed inside the perspex chamber, the inlet tube was connected to an ⁸¹Rb/^81mKr generator, and the outlet valve was left open to air.

Six dual-energy ventilation and perfusion simultaneous SPECT acquisition protocols were acquired for comparison with varying collimators, matrix sizes, number of angles of acquisition, and time per acquisition, as shown in Table 1, using the Siemens Symbia T16 SPECT/CT scanner (Siemens Medical Solutions USA). Two energy windows were set with a 15% width, centred around 140 keV for ^99mTc and 190 keV for ^81mKr to minimise the downscatter of ^81mKr into the ^99mTc window.

Table 1.

Six different acquisition protocols were investigated with a range of different choices for collimator, matrix size, number of views, and time

Acquisition	Collimator	Matrix	Views	Acquisition duration per view(sec)	Total duration (mins)
Protocol 1	MEGP	128	64	15	14.4
Protocol 2	MEGP	64	64	15	14.4
Protocol 3	LEAP	64	64	15	14.4
Protocol 4	LEAP	128	128	8	14.9
Protocol 5	LEHR	64	64	15	14.4
Protocol 6	LEHR	128	128	8	14.9

Three different collimators were used for repeated acquisitions, as detailed in Table 2. The time per view was adjusted to ensure that the total SPECT imaging duration was kept constant between the different protocols. The same image reconstruction protocol was used in all images using the T16 Siemens Symbia scanner reconstruction software (Siemens Healthineers, Germany) with 2D-OSEM (Ordered Subsets Expectation Maximisation) iterative algorithm in 4 iterations, 16 subsets, and a 1 cm full-width half-maximum Gaussian filter.

Table 2.

Collimator specifications

Collimator	Relative sensitivity (to LEHR)	Quoted system resolution at 10 cm(mm)	Septal thickness(mm)	Septal penetration at 190 keV (%)
LEHR	1	8.4	0.16	13.8
MEGP	1.53	12.5	1.14	0.02
LEAP	1.63	9.4	0.2	14.9

A dedicated SPECT viewing and volumetric quantification platform (Hermes Hybrid viewer 5.1.1-Hermes Medical Solutions Sweden) was utilised for the visualisation of ventilation, perfusion, and V/Q quotient images (Fig. 2). Only the large cylindrical defect was seen in all images. In the V/Q quotient image, the mismatched defect was seen mainly in orange, while the matched defect was seen in black, corresponding to the higher and lower ends of the image colour scale, respectively (Fig. 2c).

Fig. 2.

Square shaped phantom visualised as the perfusion image (A), ventilation image (B), V/Q quotient image (C)

The perfusion defect quantification was performed using the V/Q quotient image [16]. The V/Q quotient image contained successive count index thresholds of 0 to 6, correlating to the colour scale of the image, which could be utilised for partitioning and quantifying a volume of interest (VOI) [16] (Fig. 2). The count index thresholds 1–5 were applied to the mismatched defect in the axial plane, in succession, to identify the optimum threshold quantifying the best matched VOI to the actual defect volume of the larger cylinder of 70 cm [3]. The extreme thresholds (0,6) were not analysed as they would include all and none of the counts, respectively. This process of quantification was carried out for all 6 acquisition protocols, and the optimum threshold with its corresponding defect volume for the actual volume for each acquisition protocol was identified.

2. Identification of optimum V/Q reconstruction parameters and further evaluation of simultaneous dual acquisition accuracy for PE quantification using the Data Spectrum anthropomorphic phantom

This phase of the study aimed to build on the experience of the previous square phantom work to create a more realistic PE phantom using an anthropomorphic torso phantom (Data Spectrum Co, USA) [17]. The.

same type of specially sourced sponge as before was used to create a 270 cm [3] volume cylindrical sponge containing a 70 cm³ conical central defect to more accurately represent the configuration of a typical PE. This sponge was placed inside the lung cavity of the torso phantom to create photon attenuation and scattering conditions from surrounding body tissues (Fig. 3).

Fig. 3.

Cylindrical sponge phantom within the Data spectrum torso phantom

The sponge was drenched in 200 MBq 99mTcMAA, drip-dried, and inserted into a thin plastic container made just large enough for the sponge to be inserted into it before it was introduced into the larger lung cavity of the torso phantom (Fig. 3). After placing the phantom on the imaging couch, the circular opening was sealed off except for a small opening where the tube supplying ^81mKr gas to the cylindrical sponge was inserted.

Imaging was performed using the Siemens Symbia T16 scanner (Siemens Healthineers, Germany) with the departmental clinical dual tracer simultaneous acquisition SPECT/CT protocol of MEGP collimator, 64 projections, 15 s time per view, 128 square matrix size, 15% energy window, and 190 keV peak for ^81mKr and 140 keV window for ^99mTc. Once the ^81mKr gas generator turned on, simultaneous ventilation and perfusion SPECT acquisitions were obtained.

Images were reconstructed with dedicated SPECT image reconstruction software (Hermes Hybrid recon 3.2.1- Hermes Medical Solutions Sweden) using both OSEM and filtered back-projection (FBP). In OSEM different sub-iterations (2, 4, 8, 16, 32, 64, 128, 256, 512, 1024) with different full-width half-maximum Gaussian filters (0, 1 cm, 2 cm) summing to 30 parameter combination groups and in FBP Butterworth cut off 1,2 with order 5, 10, 15, 20 summing to 8 reconstruction parameter combination groups were utilised for assessment. Our departmental clinical reconstruction protocol of 2D-OSEM 4 iterations, 16 subsets, and 1 cm full-width half-maximum Gaussian filter was also included in this assessment to allow comparison with the current departmental practice.

The resulting datasets for each reconstruction protocol were analysed using a dedicated SPECT viewing and volumetric quantification platform (Hermes Hybrid viewer 5.1.1-Hermes Medical Solutions Sweden) utilising ventilation, perfusion, and V/Q quotient images.

In both ventilation and simultaneous perfusion image datasets, 5%, 10%, 15%, 20%, 25%, 30%, 35%, 40%, 45%, and 50% of the maximum count intensity were used as thresholds for automated image volume quantification. For this volumetric quantification, first, the absolute counts within a given %count intensity threshold of a volume of interest (VOI) were calculated as a percentage of the total counts in the image. Then, by applying this result to the volumetric software, the volume of the VOI within the % count intensity threshold concern was computed automatically in cm³.

The V/Q quotient images were quantified using the 1, 2, 3, 4, and 5 count index threshold values for automated volumetric quantification. The final quantification results were compared to the physical sponge and defect volumes to identify the optimum count thresholds with the most compatible defect volume.

3. Application of phantom-based semi-automated V/Q SPECT pulmonary embolism quantification parameters and evaluation of %PE quantification methods using clinical datasets

This phase of the study aimed to establish which of the two semiautomated %PE quantification approaches most closely aligned to a detailed manual ‘Gold standard’ delineation of %PE, the volumetrically derived data method (VDD) using ventilation and perfusion images, or the quotient image data method (QID).

Quantification was performed using VDD and QID methods, first using phantom optimised reconstruction (POR) parameters, and then, for completeness, the departmental clinical reconstruction (DCR) parameters were also analysed.

Patient study selection

100 consecutive PE-positive V/Q SPECT/CT patient studies from 30/09/2019 backward (pre-COVID-19 era) were retrospectively selected from the radiology image system of our centre. Inclusion criteria consisted of a radiologist diagnosed PE-positive study report, use of ^81mKr as the ventilatory radiopharmaceutical, a simultaneous acquisition protocol, and completion of all ventilation, perfusion, and low-dose CT stages of the study. Exclusion criteria were studies with motion artifacts and indeterminate or PE-negative reports from the reporting radiologist.

Acquisition and reconstruction protocol

Perfusion and ventilation imaging were carried out with the use of ^99mTc MAA 200MBq and ^81mKr 40- 400MBq, respectively, with an MEGP collimator, 128 projections, 64 square matrix size, with a 15% energy window used as the acquisition parameters for both radioisotopes. V/Q SPECT simultaneous acquisition was followed by a 19 mAs 110 kV low-dose CT with 1.5 mm slice thickness. Two image datasets were reconstructed from raw processing data using the POR protocol with OSEM 8 iterations, 16 subsets, no filtering for ventilation and 16 iterations, 16 subsets, no post-reconstruction filtering for perfusion and the DCR protocol with OSEM 4 iterations, 16 subsets, and post-reconstruction 1 cm Gaussian filter with dedicated SPECT reconstruction and viewing software (Hermes Hybrid recon 3.2.1- Hermes Medical Solutions Sweden).

Image evaluation and quantification

The images were visualised and quantified using a dedicated SPECT viewing and volumetric quantification system (Hermes Hybrid viewer 5.1.1-Hermes Medical Solutions Sweden). The quantification of total %PE was performed using the 3 methods described below.

1. The gold standard (GS) manual method of %PE quantification

For each study, a detailed manual GS %PE was derived utilising the SPECT/CT co-registered hybrid image from the POR reconstruction protocol using a dedicated SPECT viewing and quantifying platform (Hermes Hybrid viewer 5.1.1-Hermes Medical Solutions Sweden) (Figs. 4 and 5) as per the following steps. Firstly, in the perfusion SPECT/CT hybrid image, mismatched perfusion defects were identified. The low-dose CT was used to rule out any other underlying pathology as the cause of the mismatch defect. ROIs were drawn manually, slice by slice, in each trans axial slice using the CT for identification of lung margins when possible(Fig. 4). Then, for each mismatched perfusion defect, all the manually drawn ROIs from all trans-axial slices were combined to form a VOI using the volumetric quantification software (Hermes Hybrid viewer 5.1.1-Hermes Medical Solutions Sweden). The total PE volume was calculated by the addition of all the mismatched perfusion defect volumes in the study.

Fig. 4.

Manually drawn VOI in the perfusion/CT image

Fig. 5.

Manually drawn VOI in the ventilation/CT image

Secondly, using the ventilation SPECT/CT hybrid image, the ventilation volumes were identified, and ROIs were drawn manually in trans axial slices by slice in each lung using the edge of the CT ventilation image for identification of lung margins (Fig. 5). Then, all the manually drawn ventilation ROIs from all trans-axial slices were combined to form a total ventilation volume of a lung using the volumetric quantification software (Hermes Hybrid viewer 5.1.1-Hermes Medical Solutions Sweden). The total lung volume was calculated by the addition of the two lung volumes.

Finally, the total %PE was calculated as per the formula below.

To derive this GS % PE quantification, around 60–150 min were spent on each V/Q SPECT/CT scan.

• 2.Volumetrically derived data (VDD) method of %PE quantification

The ventilation, perfusion image volumetrically derived data (VDD) %PE quantification method is described below:

To obtain the perfusion and ventilation volumes, % count intensity thresholds derived from the phantom work were applied to the volumetric quantification software (Hermes Hybrid viewer 5.1.1-Hermes Medical Solutions Sweden) separately for each of the two reconstruction protocols.

For the images from the POR protocol, 25% for ventilation and 22% for perfusion % count intensity thresholds were applied. Some studies consisted of ventilation images with a higher level of ^81mKr remaining in the upper airways as compared to the lung parenchyma and were masked before the application of the cutoff threshold. In each image, the VOI resulting from the voxel contents above the applied thresholds was visually analysed for accurate delineation of the activity within lung volumes, and marginal alterations to the threshold had to be made at times to ensure the VOI conformed to all the activity within the visualised lungs (Fig. 6). The lung perfusion volumes were derived similarly (Fig. 7). The total perfusion defect volume was calculated by subtracting the total perfusion volume (of both lungs) from the total ventilation volume (of both lungs). Finally, the total %PE was calculated per the same formula used for the gold standard method.

Fig. 6.

Ventilation VOI

Fig. 7.

Perfusion VOI

For calculating the %PE in the images from the DCR protocol, the same procedure as for the POR protocol was followed, except the count intensity thresholds for ventilation and perfusion were 27% and 23%, respectively, as per the phantom work described previously.

• 3.V/Q quotient image data (QID) method of %PE quantification

For the V/Q quotient image data (QID) method of %PE quantification, first, the total PE volume was derived by quantifying and adding each wedge-shaped mismatched PE defect VOI highlighted within the Hermes V/Q quotient image with a count intensity threshold of 2 (Fig. 8). The total ventilation volume was derived in the same way as in the VDD method. The total %PE was calculated using the same formula as used for the gold standard method.

Fig. 8.

Mismatched (PE) defect VOI

To summarize the PE quantification methodology of this section, each patient’s SPECT V/Q datasets were reconstructed using reconstruction protocols (POR and DCR). The reconstructed datasets from both protocols were then quantified to calculate the %PE using VDD, QID quantification methods, and only the POR protocol image was used in the gold standard manual %PE quantification method.

The results of POR/VDD, POR/QID, and DCR/VDD, DCR/QID %PEs were then compared to the gold standard results. The results of this analysis were displayed in scatter and Bland-Altman plots and statistically analysed to assess which V/Q SPECT PE quantification method most closely aligned with the gold standard results.

Results

1. Identification of optimum V/Q acquisition parameters using the Leeds Test Lung phantom

The use of protocol 6 (LEHR collimator, 128 matrix, 128 views) with the count index threshold 2 resulted in the exact equivalence of the measured 70 cm [3] volume of the defect, making this the most accurate acquisition protocol for perfusion defect quantification (within the specific simulated test object conditions) (Fig. 9). The use of acquisition protocol 1, which is the routine departmental clinical protocol (MEGP collimator, 128 matrix, 64 views) with count index threshold 2, resulted in a measured volume of 74 cm [3] (Fig. 9), showing only a 6% overestimation of the actual volume, and was identified as the second most accurate protocol for image quantification. This demonstrates that the clinically used departmental protocol caters to simultaneously acquiring 99mTc and 81mKr with the MEGP collimators and offers some advantages in terms of handling noise in the data when compared to most other acquisition protocols.

Fig. 9.

Defect volume vs. count index threshold in varying acquisition protocols (True volume is marked in a red horizontal line)

Due to the constraints of the COVID-19 pandemic on a prospective study and as the second most accurate acquisition protocol, the routine departmental clinical protocol, which acquired retrospective clinical cases, was deemed acceptable for further quantitative assessment of PE work in this study.

• 2.Identification of optimum V/Q reconstruction parameters and further evaluation of simultaneous dual acquisition accuracy for PE quantification using the Data Spectrum anthropomorphic phantom

OSEM reconstruction

Ventilation images

In ventilation images, all OSEM reconstruction parameter groups had a corresponding volume of < 5% relative error compared to the true volume at some % count intensity threshold (ranging from 21 to 45% according to the reconstruction parameter combination used) when thresholds were tested sequentially. This included the current departmental clinical protocol with a 27% count intensity threshold, resulting in an image quantification volume with < 5% relative error. Thus, it was not possible to choose an optimum reconstruction parameter combination group considering the volume measurements alone, and the image with the best matching shape to the physical shape of the phantom was considered. The shape of the image was matched to the cylindrical shape of the phantom by considering if all the axial slices’ region of interest (ROI) was a symmetrical circle of equal size, with fairly regular ROI boundaries.

Increasing the number of sub-iterations increased the image resolution and accuracy of the image. However, it also resulted in jagged/irregular ROI boundaries, most likely due to the intensified heterogeneity of tracer distribution distorting the image. The addition of post-reconstruction filtering also decreased the image resolution and accuracy. Thus, a compromise was found in the mid-ranged sub-iterations of 128 (8 iterations, 16 subsets) with no filtering determined as the optimum reconstruction parameter group with a count threshold of 25% (with 0.4% percentage error) for ventilation images. (Fig. 10)

Fig. 10.

Optimum ventilation protocol image: 8 iterations, 16 subsets with no post-reconstruction filtering, showing the circular shape of the phantom in the trans-axial plane

Perfusion images

In simultaneous perfusion images, almost all (except for excessively high sub-iterations of I 32 S 32 GF = 0 in sequential acquisition) OSEM reconstruction parameter groups had an optimum % count intensity threshold (ranging from 16 to 43%) with a corresponding volume of < 5% relative error compared to the true volume. This included the current departmental clinical protocol, which displayed an optimum count intensity threshold of 23%. Thus, it was not possible to choose an optimum reconstruction parameter combination group considering the volume alone, and a morphological assessment of axial slice ROIs was carried out, compared to the physical defect shape of the phantom, to determine the optimum reconstruction parameter group.

Increasing the number of sub-iterations increased the image resolution and accuracy of the shape of the image. However, it also resulted in jagged/irregular ROI boundaries, most likely due to the heterogeneity of tracer distribution, resulting in some degree of image distortion. The addition of post-reconstruction filtering also decreased the image resolution and accuracy.

Fig. 11.

Optimum perfusion protocol image: 16 iterations, 16 subsets, and no post-reconstruction filtering showing the ‘Phantom PE’ defect. Subsequently, 16 iterations and 16 subsets with no post-reconstruction filtering were determined to be the most compatible reconstruction parameters for simultaneous perfusion images with a 22% optimum count intensity threshold, (with 0.1% percentage error), (Fig. 11)

V/Q quotient images

All the ventilation image reconstruction parameter combinations with the corresponding simultaneous perfusion image reconstruction parameter combinations were then used for the construction of the V/Q quotient image. Additionally, the combination of optimum ventilation parameters of 8 iterations, 16 subsets GF, and optimum simultaneous perfusion parameters of 16 iterations, 16 subsets with no filtering was also used for this assessment.

The most accurately quantified V/Q quotient image resulted from the combination of the optimum ventilation and perfusion parameter combinations, as when V/Q count index threshold 2 was used, this image displayed the best fitting shape to the physical defect shape, with an image defect volume was 69.47cm [3] (0.76% percentage error) (Fig. 12).

Fig. 12.

Optimum reconstruction parameter (ventilation I8 S16 GF = 0, simultaneous perfusion I16 S16 GF = 0) V/Q image

Filtered back-projection

All image reconstructions with filter back projection were distorted by the star artifact in varying degrees, resulting in poor image quality compared to iterative reconstruction, so this method was not applied.

• 3.Application of phantom-based semi-automated V/Q SPECT pulmonary embolism quantification parameters and evaluation of %PE quantification methods in clinical datasets

The POR protocol images (Figs. 13, 14, 15 and 16) showed a better correlation and lower standard deviation (SD) when compared to the DCR protocol images. The POR images derived VDD and QID methods resulted in R² = 0.95 and R² = 0.96, while DCR images derived VDD and QID methods showed R² = 0.84 and R² = 0.91. Additionally, POR protocol images displayed bias and standard deviation (SD) of 1% and 3% for both VDD and QID methods, and DCR images derived from VDD and QID methods showed 1% bias for both methods and 5% and 6% SD for QID and VDD methods, respectively.

Fig. 13.

POR QID %PE vs. gold standard %PE

Fig. 14.

POR VDD %PE vs. Gold standard %PE

Fig. 15.

%PE mean vs. %PE difference in POR QID and gold standard

Fig. 16.

%PE mean vs. %PE difference in POR VDD and gold standard

The average time to perform the QID method in a single study was measured to be around 3–5 min, while 5–10 min were taken for the VDD by an experienced operator.

Thus, the POR/QID %PE quantification method (Figs. 13 and 15) showed a better fit with the gold standard, and faster performance time as such is our chosen method for semiautomated SPECT V/Q %PE quantification.

Discussion

It was evident from the work of this study that volumetric V/Q SPECT results were dependent on scan acquisition and reconstruction parameters and that the ideal parameters for quantification differed from the optimum parameters for current clinical visual reporting.

The ideal acquisition parameters in volumetric PE quantification V/Q SPECT required an LEHR collimator with 128 views compared to the routine visual reporting requirements of an MEGP collimator with 64 views. This demonstrates the requirement for a higher resolution for quantification and moderate noise for visual reporting. Volumetric quantification seems to be more reliant on superior contrast with low low-energy collimator with 81mKr use and an increased number of views.

The anthropomorphic phantom work demonstrated the effect of varying reconstruction parameters on V/Q SPECT volumetric quantification. It also showed that the optimal % count intensity threshold for quantification differed according to the reconstruction protocol.

The optimum reconstruction parameters identified by this study differed from our departmental clinical V/Q SPECT reconstruction protocol for visual reporting. Optimised image reconstruction established using phantoms found volumetric quantification warranted a higher number of sub-iterations (128 for ventilation images and 256 for perfusion images compared to 64 sub-iterations in visual reporting) and no post-reconstruction filtering, contrasting visual reporting requirements. The difference in sub-iterations in ventilation and perfusion images could be explained by the need for higher resolution in perfusion images to display the small photopenic defects accurately when compared to ventilation. Similar to the quantification requirements identified in the acquisition protocol, the increased number of sub-iterations resulted in higher resolution images, and no post-reconstruction filtering resulted in improved contrast, likely at the expense of signal-to-noise levels. However, at exceedingly high sub-iterations, threshold ROI (region of interest) outlines became irregular, most likely due to intensified reconstructed tracer heterogeneity.

Palla et al. [5] in 2001 identified a count intensity cut off of 25% for V/Q SPECT perfusion images based on their phantom work, but the reconstruction protocol used filter back projection (Hanning filter, 0.8 cycles/cm cut-off frequency). Another group has recommended a V/Q SPECT perfusion image (OSEM: 16 sub-iteration and Butterworth post filtering) cutoff threshold of 40% based on the optimum correlation with the mean pulmonary arterial pressure of 30 patients [14]. However, as demonstrated above, the cutoff thresholds are dependent on the reconstruction parameters; thus, a direct comparison with the results of our study with these previous studies was not possible.

To our knowledge, this work is the first to assess the effects of acquisition and reconstruction parameters on PE clot burden quantification using V/Q SPECT datasets.

This work also demonstrated that dual isotope simultaneous V/Q SPECT imaging can be successfully used in the volumetric quantification of lung volumes utilising cutoff thresholds. However, due to the probable downscatter effect of 81 m Kr into the 99 m Tc window, it might be prudent to use a scan performed by the same mode of acquisition (sequential/simultaneous) when comparing %PE between scans.

In terms of the PE quantification method, the VDD(volumetrically derived data) method required a sliding scale adjustment of up to +/- 5% of the phantom-derived % count intensity thresholds for accurate volume delineation on a case-by-case basis. This is in line with Bourhis et al. [18], which stated the physiological nonuniformity of V/Q radioisotope distribution among individuals.

The QID method showed a marginally better correlation when compared to the gold standard method than the VDD method and did not require any individual tailoring of threshold during the process of obtaining mismatched defect volume (PE volume) using the V/Q quotient image, leading to less time consumption and less intra/inter operator variation. It is anticipated that in the future, AI (Artificial Intelligence) techniques may help achieve a method for %PE calculation from V/Q SPECT datasets, which will require even less user input.

One of the limitations of this work was that the option for CT-based attenuation correction (AC) and scatter correction (SC) was not explored. This was mainly due to the lack of use of AC and SC in routine clinical reporting and the lack of validation of AC/SC in the dual isotope setting. Due to other priorities and time constraints, it was not possible to explore these in the current study. Consideration of AC and SC as part of iterative reconstruction may not only improve image quantification in general but may potentially improve other image quality indices, such as improving the signal-to-noise aspect of OSEM. This is a direction that may be explored in the future.

Other parameters not considered in this study include the role of patient respiratory motion (and potentially misregistration of image phases), which was not simulated in the phantom study due to various time constraints and facility access limitations during the time of the project. Although authors have previously speculated on such effects [19] a systematic study of the influence of respiratory motion could add further understanding of the methodology limitations, next to those of spatial resolution, and it should be considered in any future work.

Conclusions

The V/Q SPECT volumetric quantification results of lung defect volumes are dependent on acquisition, reconstruction parameters, and cutoff thresholds. Accurate volumetric clot burden quantification can be achieved with simultaneous dual tracer acquisition. This work identified, evaluated, and clinically validated the optimum and logistically most feasible parameters for a semiautomated volumetric V/Q SPECT %PE quantification method. The next phase of the study will be to apply this method to clinical cases to allow a comparison of %PE with patient acute and chronic clinical parameters, and to assess whether quantification can be utilised in improving patient management and outcome prediction in PE.

Abbreviations

AC

Attenuation correction

CT

Computed tomography

CTPA

Computed tomography pulmonary angiogram

DCR

Department of clinical reconstruction

EANM

European association of nuclear medicine

FBP

Filter back projection

GF

Gaussian filter

I

Iterations

LEAP

Low-energy all-purpose

LEHR

Low energy, high resolution

MAA

Macroaggregated albumin

MEGP

Medium energy, general purpose

OSEM

Ordered subset expectation maximization

PE

Pulmonary embolism

POR

Phantom optimised reconstruction

QID

Quotient image data

ROI

Region of interest

S

Subsets

SC

Scatter correction

SD

Standard deviation

SPECT

Single photon emission computed tomography

V/Q

Ventilation/Perfusion

VDD

Volumetrically derived data

VOI

Volume of interest

Author contributions

HF designed and performed the study, performed the analysis, and wrote the manuscript. CSA designed and performed the first section (analysis of acquisition parameters) of the study. CH and LL contributed to the design, contributed resources, and checked the manuscript. All authors read and approved the final manuscript.

Funding

This research was conducted without any external funding.

Data availability

The datasets used and/or analysed during the current study are available from the corresponding author on reasonable request.

Declarations

Ethics approval and consent to participate

The Guys and St Thomas’ Trust CLIMP (Clinical Information Management Programme) committee was informed of the project. Due to the retrospective nature, informed consent was waived.

Consent for publication

Not applicable.

Competing interests

The authors declare that they have no competing interests.