Retrospective fractional dose reduction in Tc-99m cardiac perfusion SPECT/CT patients: A human and model observer study

Abstract

Background:

In the ongoing efforts to reduce cardiac perfusion dose (injected radioactivity) for conventional SPECT/CT systems, we performed a human observer study to confirm our clinical model observer findings that iterative reconstruction employing OSEM (ordered-subset expectation-maximization) at 25% of the full dose (quarter-dose) has a similar performance for detection of hybrid cardiac perfusion defects as FBP at full dose.

Methods:

One hundred and sixty-six patients who underwent routine rest-stress Tc-99m sestamibi cardiac perfusion SPECT/CT imaging and clinically read as normally perfused, were included in the study. Ground truth was established by the normal read and the insertion of hybrid defects. In addition to the reconstruction of the 25% of full dose data using OSEM with attenuation (AC), scatter (SC), and spatial resolution correction (RC), FBP and OSEM (with AC, SC and RC) both at full dose (100%) were done. Both human observer and clinical model observer confidence scores were obtained to generate receiver operating characteristics (ROC) curves in a task-based image quality assessment.

Results:

Average human observer AUC (area under the ROC curve) values of 0.725, 0.876, and 0.890 were obtained for FBP at full dose, OSEM at 25% of full dose, and OSEM at full dose, respectively. Both OSEM strategies were significantly better than FBP with p-values of 0.003 and 0.01 respectively, while no significant difference was recorded between OSEM methods (p=0.48). The clinical model observer results were 0.791, 0.822, and 0.879, respectively for the same patient cases and processing strategies used in the human observer study.

Conclusions:

Cardiac perfusion SPECT/CT using OSEM reconstruction at 25% of full dose have AUCs larger than FBP and closer to those of full-dose OSEM when read by human observers, potentially replacing the higher dose studies during clinical reading.

INTRODUCTION

The United States and Canada have the highest annual number of myocardial perfusion imaging (MPI) exams performed in the world, with the United States performing approximately twice the number as in Canada (2500/million vs 1200/million) (1). Using the linear no-threshold (LNT) model for radiation-induced cancer risk, Berrington de Gonzalez et al (2) estimated that approximately 7400 new cancers will occur in the United States annually due to radiation exposure attributed to MPI. Although opposing views do exist regarding the validity of the LNT model (3, 4), it always prudent to err on the safe side. Furthermore, a recent worldwide study (5) under the auspices of the International Atomic Energy Agency (IAEA) identified 8 parameters to use as a "best practices quality index (QI)" to evaluate participating laboratories in their quest to reduce the overall dose. The higher the adherence to these best practice QI parameters the better, with a score of ≥ 6 the goal. North America had the lowest average score of 4.7. Therefore, it is imperative to improve standard clinical practice (6) and reduce the amount of radioactivity injected, and subsequent exposure to the potential harmful effects of radiation (7-19). Ourselves and others work diligently to implement these improvements on the most widely available type of SPECT systems (7-12, 14, 17, 18). Many studies have investigated the usage of iterative reconstruction methods using ordered-subset expectation-maximization (OSEM) implementations (20) with distance-dependent resolution compensation (RC) (8-12, 15), while several have endeavored to also correct the influence of attenuation (AC) and scatter (SC) (7, 14, 17). Some of these studies showed the feasibility of either reducing the time of the MPI acquisition (7-10), the injected radioactivity (9, 11, 12, 16, 17), or doing stress only imaging (7, 18, 19, 21) employing conventional SPECT cameras. Moreover, all the image and processing strategies described above are also included in the IAEA QI (5).

In our recent work (17), we optimized a number of processing strategies for different levels of dose reduction using the area under the curve (AUC) values from receiver operator characteristic (ROC) curves. This was accomplished by employing the total perfusion deficit (TPD) score the QPS (quantitative perfusion SPECT) software developed at Cedars Sinai (22) as a clinical model observer for perfusion-defect detection. The aim of this work was to confirm the clinical model observer results from our recent study (17) employing human observers. More specifically, we wanted to determine if OSEM with AC, SC, and RC at a 25% dose level was comparable in defect detection to filtered back-projection (FBP) at full dose, when the processing parameters of the reconstruction methods were optimized for the defect detection task using TPD score as a clinical model observer (17).

METHODS AND MATERIALS

Patient Group

From a pool of 1250 patient studies (633 female) acquired with written consent approved by the institutional review board of the University of Massachusetts Medical School, the perfusion SPECT/CT data of 166 patients (87 female) were selected for inclusion in this investigation on the basis of a normal cardiac perfusion SPECT outcome as reported by the attending cardiologist or nuclear medicine physician. All patients, except one with only a stress acquisition, underwent same-day rest-stress Tc-99m sestamibi cardiac-perfusion SPECT/CT imaging employing a two-headed Philips BrightView SPECT/CT (Cleveland, Ohio) with the heads in the 90-degree configuration. In addition to the clinical frame-mode acquisitions through 180 degrees acquired into 64 projections from right-anterior-oblique (RAO) to left-posterior-oblique (LPO), list-mode data were also obtained. A cone-beam CT acquisition was done prior to the SPECT while the patients breathed normally. Only the stress acquisitions with a duration of 10.67 minutes (20 sec per projection) were utilized. Both physical (n=99) and pharmacological (n=67) stressed patients were included. ASNC guidelines were used to determine radioactivity to inject (23) with 10-12 mCi for rest and 30-36 mCi for the subsequent stress, depending on the BMI. Unit doses from a contracted radiopharmacy was used and meticulous records of actual injected radioactivity amounts, including syringe retention, were kept. To further quantify the effect of BMI on injected radioactivity, we also kept track of total acquired projection counts of all patients.

The list-mode data were used to form two frame-mode data sets with 15% energy windows straddling the photopeak of Tc-99m (140.5 KeV) : 1) A set employing all the acquired counts framed into a 128×128×64 matrix with a 0.466 cm pixel size, and 2) a set with randomly selecting list-mode events throughout the acquisition, simulating a study where the injected radioactivity is only 25% of the original dose, however keeping the acquisition time, frame dimensions, and pixel size the same. In addition to these two data sets, comparable sets were formed with 4% energy windows at 121 KeV for estimation of Compton scatter employing the TEW (triple energy window) (24) method.

Patient Preparation

Of the 166 patients included in the study 83 (57 female) were used to create hybrid studies where perfusion defects were inserted as previously described (17, 25, 26). In short, the defect creation process commenced by defining a 2-dimensional (2D) region-of-interest (ROI) on a polar map generated using LV short axis slices. By reversing the process of creating the polar map from short axis slices, a short axis defect mask was defined. The short axis defect mask was reoriented back to the transverse view, superimposed on the original reconstructed transverse slices, and used to create the defect. A ratio image at each projection angle was formed by dividing a forward projection of the transverse slices with the defect present by a forward projection of the transverse slices without the defect (17, 26). Finally, the acquired projections were multiplied by the ratio images creating the hybrid projections as described in reference (26). Twenty seven (27) large, 28 moderate, and 28 small perfusion defects were simulated in 6 locations (3 single vessel defects and 3 two-vessel defects) (17, 27) and the decrease in contrast was set at 35%, 50%, and 65% compared to the local maximum at the insertion site for large and moderate defects, while the decrease in contrast was limited to 50% and 65% for the smaller defects. In a preliminary ROC study (28) where defect contrast was varied from 20% to 65% in steps 15% for all defect sizes, we found observers failed often to detect small defects with contrasts of 20% and 35% and also found it quite a challenge to recognize large and moderate defects at a 20% contrast level. These initial contrast levels were chosen to coincide with our earlier clinical model observer study design (17) where the lower contrast levels were necessary to obtain a realistic range of AUC's across all parameters evaluated.

Choice of observers

Four human observers participated in the reader study. Two of these human observers were physicists, one with extensive experience in cardiac perfusion SPECT processing and image evaluation while the other was less experienced in both, however, well grounded in SPECT technology and imaging. The other two observers were experienced cardiologists with a background in nuclear cardiology, one more experienced in the latter than the other. These choices of human observers with differing experience and background created an opportunity to investigate observers with different characteristics. We have to point out that the detection task given the human observers only required making a choice as to the presence or absence of a defect and supplying a level of confidence (related to size and severity of defect) for the choice. The decision by the human observers is not based on any clinical information (gender, history of CAD, etc) or additional imaging information (rotating planar projections, dynamic gated data display, etc.), just slices of the left ventricle. Furthermore, we have previously used this approach successfully to evaluate reconstruction strategies before proceeding to an all clinical evaluation (29-31).

Our decision to use the TPD score as a clinical model observer is related to our explanation of the human observer's task above. As in the case of human observers, the total perfusion deficit or TPD is also an attempt to calculate a metric, using a normal database, representing the confidence of having a defect present or absent (22) with higher values indicating larger and more severe defects. The TPD methodology have been shown to agree well with human observers (32).

Processing strategies for ROC comparison

In our optimization study (17), using the TPD score as a clinical model observer, optimal processing parameters were determined for a combination of dose level (injected radioactivity), reconstruction strategy, and filter parameters. Results of this optimization showed OSEM (ordered-subset expectation maximization) with AC, SC, and RC at a 25% dose level gave a similar AUC as FBP at full (100%) dose. Therefore, to test our hypothesis of not doing worse than full-dose FBP, we selected three processing strategies with parameters previously optimized for defect detection for comparison namely, 1) OSEM with AC, SC, and RC, at a 25% dose level using 16 subsets, 4 iterations, and a Gaussian post filter with a sigma of 0.559 cm (or 1.2 voxels), 2) FBP at full dose pre-filtered with a 2D Butterworth filter of order 5 and cutoff set to 0.103 cycles/cm (or 0.22 cycles/pixel), and 3) OSEM at full dose with AC, SC, and RC, using 16 subsets, 12 iterations, and a Gaussian post filter with a sigma of 0.559 cm.

ROC study design

For the purpose of establishing ground truth, we used all the reconstruction strategies evaluated herein, to visually determine uniform myocardial perfusion for patients clinically read as normal. Patients with normally perfused LV's with a history of coronary artery disease (CAD), cardiomyopathy, or having any previous interventional procedure, were not considered for inclusion. Furthermore, the insertion of a hybrid defect was also taken as establishing ground truth and in this case as having a perfusion defect present. The 166 patients included in the study were equally divided between normal (perfusion defect absent) and diseased patients (hybrid perfusion defect present) using the patients with hybrid defects inserted as described earlier. Forty (40) patients, 20 with hybrid perfusion defects absent and 20 with hybrid perfusion defects present, were used for observer training (30, 31, 33). For each of the processing strategies human observers were asked to read 50 training cases, 40 with feedback indicating defect present or not, and 10 without feedback to familiarize them with any possible difference compared to standard clinical practice (smoothness, uniformity of uptake, noise content, size and severity of defects). The last 10 cases without feedback were repeat cases from the first 40 to familiarize the human observers with the transition between training and test cases in the actual study. The observers were asked to read the remainder of the patient cases (126 cases in all, 63 with hybrid perfusion defects present) in 6 sessions, 2 sessions for each processing strategy. Each of these reading sessions commenced with 7 training cases (drawn from the 40 discussed above) to refresh the human observer's memory as to the image characteristics (smoothness, uniformity of uptake, noise content, size and severity of defects), followed by 63 test cases, for a total of 70 cases per session. All the reading sessions (training and actual sessions) were randomized for reading order (training sessions followed by the actual sessions) as well as patient case order. The human observers were asked to only complete one reading session on a given day and were not aware of the processing strategy during the time of the reading.

As in our previous human observer studies (31, 33), we employed an IDL (interactive data language) graphical user interface from ITT Visual Information Solutions (Boulder, CO) for display and scoring patient cases (Figure 1). The human observers were asked to score for absence/presence of a perfusion defect on a 5-point sliding scale (0 being normal, 4 perfusion non-existent) in 17 territories according to ASNC guidelines (23).

Figure 1:

An example of the GUI used for display and scoring using 17-segment polar map diagram (bottom center). Shown at the top are rows of short-axis, horizontal long-axis, and vertical long-axis slices. The 5-point scale and 17 territories are explained on the bottom left while feedback appears on the smaller polar map diagram on the bottom right when a training case is shown. The patient example depicts a challenging small perfusion defect in the RCA territory (yellow arrows).

For comparison, the test cases (126) were also scored using the TPD methodology (22) employing gender based normal databases created for each processing strategy. The creation of the normal databases was discussed in our optimization study (17) and as then, the normal perfused studies used for creation of the databases, were not included in the model observer study.

Analysis of observer data

Both the human observer and clinical model observer study results were analyzed using the ROCKIT software of Metz and colleagues (34, 35) to generate AUC values and the associated standard error (SE). As an adjunct, we also calculated sensitivity and specificity, with sensitivity defined as the percentage of patients with defects present correctly identified (true positive), and specificity the percentage of patients without defects correctly identified (true negative). A paired t-test was used to determine significance between the three processing strategies considered.

RESULTS

Patient demographics and projection counts

The 166 patients included in this study had an average body mass index (BMI) of 30.1 kg/m² (range 19.2-45.7) with ages ranging from 21 to 85 years (average of 59.8) and the actual average radioactivity injected was 10.1 mCi and 29.2 mCi during rest and stress respectively. Total projection counts acquired for full dose SPECT during the stress, as a result of the activity injected at stress and injected activity that remained from the rest done prior to the stress, varied between 5.3 million and 30.0 million counts (average 18.3 million ± 4.1 million). Figure 2 plots both rest and stress injected activities against BMI (top) and the total projection counts against BMI for full and 25% of full dose (bottom), respectively. The transition of higher injected radioactivity for patients with BMI's larger than 35 kg/m² are visible (Figure 2, top), though most injected activities were less than the prescribed dose due to syringe retention. Syringe retention was approximately 0.6 mCi and 2.4 mCi for rest and stress studies respectively. The outlier patient, with a BMI of 45.7 kg/m² and a recorded injected radioactivity of 19.8 mCi, shown on the stress plot (top), was the only patient where the rest and the stress were done over a two-day period. Inspecting the plotted total projection counts at full dose (Figure 2, bottom), a negative trend is barely visible, indicating that the increase in injected activity globally compensates for the increase in BMI. The patient where the rest and stress were done over two days, also records the least total projection counts. Although not immediately obvious, the total projection counts at 25% of full dose is close to an exact copy of the total counts at full dose, reduced by 75% with only the randomness of the radioactive decay adding some variation.

Figure 2:

Plots showing the distribution of rest and stress injected radioactivity doses against BMI (top) and the total projection counts of full dose and 25% of full dose (bottom) respectively.

ROC results

A summary of the ROC results is given in Table 1 and Figure 3. Table 1 lists the AUC values and SE's of individual human observers, as well as their averages and standard deviations of the averages. Table 1 also supplies the AUC values obtained using the clinical model observer. The average human observer AUC values varied from 0.732 for FBP at full dose to 0.890 for OSEM at full dose with AC, SC, and RC, while OSEM at 25% of the dose with AC, SC, and RC, fell within these limits at 0.879. Corresponding AUC values for the clinical model observer are 0.791, 0.879, and 0.822 respectively. Both OSEM strategies were statistically significantly better than FBP with p-values of 0.01 and 0.003 for OSEM at full dose and OSEM at 25% of the dose respectively, with no statistical difference recorded between OSEM at full dose and OSEM at 25% of the dose (p=0.48). Inspecting the ROC curves (Figure 3), from which the AUC values are derived, it is clear that the placement order of the processing methods for human and clinical model observers is the same, although the human observers place OSEM at 25% of the full dose with AC, SC, and RC, closer to OSEM at full dose with AC, SC, and RC, while the clinical model observer place the former closer to FBP at full dose.

Table 1:

A summary of AUC values with the standard errors in brackets.

Observer/Method	FBP 100% (SE)	OSEM 100% (SE)	OSEM 25% (SE)
Physician 1	0.745 (0.046)	0.847 (0.034)	0.859 (0.048)
Physician 2 (more experienced)	0.750 (0.041)	0.888 (0.031)	0.878 (0.032)
Physicist 1 (more experienced)	0.678 (0.041)	0.910 (0.027)	0.858 (0.035)
Physicist 2	0.754 (0.044)	0.915 (0.022)	0.920 (0.044)
Averages and standard deviations	0.732 ± 0.036 (0.043 ± 0.003)	0.890 ± 0.031 (0.029 ± 0.005)	0.879 ± 0.029 (0.040 ± 0.007)
Clinical Model Observer (TPD)	0.0.791 (0.033)	0.879 (0.030)	0.822 (0.031)

Figure 3:

Plots of the ROC results for TPD (top) and average of the human observers (bottom). The corresponding AUC values are given in Table I.

Figure 4 gives the ROC curves of individual human observers. The two ROC curves on the left depict the results of the more experienced human observers, with Physician 2 at the top and Physicist 1 at the bottom (see Table 1), while that on the right in the same order are Physician 1 and Physicist 2. As the values in Table 1 show, the more experienced human observers place the processing methods more noticeably further apart than the less experienced whom placed OSEM at full dose with AC, SC, and RC, and OSEM at 25% of full dose with AC, SC, and RC, practically at the same level.

Figure 4:

Plots of the ROC results of the individual human observers. The more experienced observers ROC curves are shown on the left with the physician at the top and the physicist at the bottom. On the right are the ROC curves of the less experienced observers in the same order. The corresponding AUC values are given in Table I.

Sensitivity and specificity results

Table 2 lists the sensitivity and specificity values for human observers and the clinical model observer. For human observers a confidence score of 0 or 1 was taken as normal perfusion or no significant decrease in myocardial uptake and a confidence score of 2-4 as a positive test for CAD or a significant decrease in myocardial uptake. The same 'confidence' cutoff value for the clinical model observer was set at 3. Although average sensitivity values for the human observers varied (75.8% for FBP, 77.4% for OSEM at full dose, and 71.8% for OSEM at 25% of the full dose), no statistical significance were present (p=0.604 between FBP and OSEM at full dose, p=0.063 between FBP and OSEM at 25% of full dose, and p=0.218 between OSEM strategies), however, the OSEM strategies have significantly higher specificity values compared to FPB (p=0.020 for OSEM at full dose, and p=0.017 for OSEM at 25% of full dose), while no difference was recorded when comparing the specificities of the OSEM processing strategies (p=0.854).

Table 2:

A summary of sensitivity and specificity values.

Observer/Method	FBP 100%		OSEM 100%		OSEM 25%
	Sensitivity	Specificity	Sensitivity	Specificity	Sensitivity	Specificity
	(%)	(%)	(%)	(%)	(%)	(%)
Physician 1	68.3	68.3	71.4	87.3	65.1	92.1
Physician 2	82.5	33.3	76.2	92.1	79.4	92.1
Physicist 1	74.6	38.1	81.0	87.3	66.7	88.9
Physicist 2	77.8	55.6	81.0	88.9	76.2	84.2
Average±Std	75.8±6.0	48.8±16.1	77.4±4.6	88.9±2.2	71.8±7.0	89.3±3.8
TPD	71.4	82.5	81.0	69.8	82.5	54.0

DISCUSSION

For the hybrid population of studies investigated herein, we found the human observers to have a lower average AUC (0.732) for the FBP full-dose studies compared to the AUC (0.791) when our clinical model observer (TPD) was applied to the same studies. The use of gender specific databases while calculating the TPD's are probably responsible for the discrepancy. The human observers only looked at the cardiac perfusion stress slices. We have previously shown that AUC values for FBP improved when given the gender and additional image information such as rotating projections, polar maps, and dynamically displayed gated SPECT images (31, 33). Sensitivity and specificity results of human observers and the clinical observer (TPD) for FBP confirm the abovementioned reasoning with humans recording slightly better average sensitivity than the clinical observer (75.8%±6.0% versus 71.4%), while recording much worse specificities (48.8%±16.1% versus 82.5%). Keep in mind that the AUC is a composite of sensitivity and specificity, hence AUC values of 0.731 and 0.791 for humans and TPD respectively. Comparing FBP with the OSEM processing strategies for human observers, AUC values clearly placed both OSEM at full dose and OSEM at 25% of the dose above FBP. This placement order is also true when comparing human observers' specificity results, however sensitivity results are less promising with no statistically significant difference between the processing strategies. Two reasons are probable, first, the wide range of sizes and contrast of inserted defects are challenging, and secondly, the erroneous identification of breast and diaphragmatic attenuation as defects in FBP (see examples below). The discrepancy in specificity values for the OSEM processing strategies between human observers and the clinical model observer are less obvious, however, careful inspection of the TPD scores (clinical model observer) for normal test subjects (true negatives), from which specificity is calculated, reveal a number of unusually high values compared to FBP. These high values mainly belonged to normal test subjects with prominent outflow tracks (shorter septal walls) and all the corrections applied during OSEM processing, improved resolution and eliminate wall deformation present in FBP data due to the 180-degree acquisition. The TPD methodology seems to be unable to account for this anatomic reality. Note that the AUC’s for the clinical model observer in the present study are higher than in our optimization study (17) because of the elimination of the lower contrast defects as described in the Materials and Methods section.

To illustrate the reason for the discrepancy between FBP and the OSEM processing strategies, three example cases are presented (Figure 5-7). In the first (Figure 5) a male with normal perfusion is shown. FBP clearly shows an attenuation artifact in the inferior wall while both OSEM methods show normal perfusion. All the human observers read FBP as diseased and the OSEM methods as normal. TPD also scored FBP as having normal perfusion. In the next patient example (Figure 6) a moderate sized perfusion defect was inserted into a female patient heart across the RCA-LCX boundary with a 50% contrast level. The patient happened to have a significant attenuation artifact anteriorly due to breast attenuation and all the human observers incorrectly identified this artifact as a defect in FBP while correctly identifying the defect across the RCA-LCX boundary with the two OSEM processing strategies. However, in ROC methodology where location is not considered, all observers still correctly identified the patient as having coronary artery disease. TPD scoring was also successful in diagnosing the patient as having coronary artery disease; however, it should be noted that this metric (TPD) is a global measure. Thus, when inspecting the individual scores (6.1, 4.6, and 4.6 for FBP, OSEM at full dose, and OSEM at 25% of full dose, respectively), it seems as if FBP probably used both the anterior and inferior decreases in counts to arrive at the score. Finally, the next patient example (Figure 7) is a male with an attenuation artifact in the inferior portion of the left ventricle due to enhanced attenuation. A large perfusion defect was inserted in the LCX territory at a 50% contrast level enhancing the attenuation artifact noticeably in FBP compared to the OSEM methods. The human observers were successful in identifying the defect in all the processing strategies, however, rated FBP full-dose and OSEM at a 25% dose level with AC, RC, and SC, with higher confidence compared to OSEM at full dose with AC, RC, and SC. Also, the clinical model observer (TPD scoring), concur with the human observers and rate both FBP and OSEM at 25% of the full dose at a higher confidence level than OSEM at full dose (4.1, 3.2, and 4.9 for FBP, OSEM at full dose, and OSEM at 25% of full dose, respectively). In some cases, the humans incorrectly flagged larger areas apparently affected by the disease when reading FBP. The reason for the increased confidence in the rating for FBP is obvious, however the increased confidence rating for OSEM at 25% of the dose with AC, RC, and SC, is less obvious. We suggest two possible reasons. The first is the effect convergence will have on the decreased activity embedded in a hot background with a different number of iterations for OSEM at a 25% dose level with AC, RC, and SC, compared to OSEM at a full dose level with AC, RC, and SC, (4 versus 12 iterations). Secondly, the effect of increased noise for OSEM at a 25% dose level with AC, RC, and SC, due to the decrease in count level enhanced by the diaphragmatic attenuation. Although the last two examples, have a minimal influence on the AUC values for FBP (all observers correctly identified CAD), many other patient cases without perfusion defects inserted have similar attenuation artifacts as that shown in Figure 5, leading to false positive readings and lower AUC values. For, OSEM at a 25% dose level with AC, RC, and SC, the difference in convergence rate and noise structure leads to better AUC values, however with a larger variation between the human observers, seemingly depended on the level of expertise. Furthermore, the human observers were only given the non-gated stress slices for their decision-making process, while the clinical model observer were given gender based normal databases tailored to each processing method. In a previous ROC study (33) we showed that FBP with all imaging information (stress and rest perfusion slices, rotating projections, dynamically displayed gated data, and polar maps) and gender given to the human observers resulted in improved AUC values compared to employing only stress slices (31).

Figure 5:

An example of a male patient (age 61, BMI 25.2 kg/m²) with normal perfusion. At the top (a) representative transverse slices of the attenuation map spanning the inferior portion of the heart, followed by FBP (b), OSEM full dose (c), and OSEM 25% of full dose. The effect of the lung-diaphragm attenuation transition is clearly visible (indicated by the yellow arrows) on FBP. The two OSEM reconstruction strategies looks similar with the reduced dose (25%) smoother in appearance due to the difference in optimization (17).

Figure 7:

A male patient example (age 64, BMI 29.8 kg/m²), with a large hybrid defect inserted at a 50% contrast level in the LCX territory. From top to bottom (a) representative transverse slices of the attenuation map spanning the inferior portion of the heart, (b) FPB, (c) OSEM full dose, and (d) OSEM with 25% of full dose. For each of these processing strategies, the short axis, horizontal long axis, and vertical long axis views are given. The arrows are added to show the location of the inserted defect. In the case of the vertical and horizontal long axes the arrows only delineate the visible defect range, while the arrows point to the defect at every alternative short axis slice.

Figure 6:

An example of a female patient (age 62, BMI 32.2 kg/m²) with a moderate sized inserted defect at a 50% contrast level across the RCA-LCX boundary (indicated by the arrows). At the top (a) representative transverse slices of the attenuation map spanning the anterior portion of the heart, followed by (b) FBP short, horizontal and vertical long axis slices with the effects of breast attenuation clearly visible anteriorly, and (c) full dose OSEM with the attenuation artifact corrected and the defect clearly visible. At the bottom (d) 25% of full dose OSEM is given, appearing to be similar than full dose OSEM except for more smoothing.

Our results confirmed and strengthen the previous work done in the field of dose reduction in MPI (7, 9, 11, 15, 16), however, due to approach and protocol differences a direct comparison is not possible. Our results can also be an incentive to reduce acquisition time to one quarter (quarter-time) (8) instead of reducing the injected dose or reducing the injected dose by half as well as reducing the acquisition time by half (half-time, half-dose), with the latter the more prudent and in line with the IAEA QI (5). Furthermore, when half-time stress only imaging (7) is combined with half-dose imaging in the majority patients (21), significant radiation exposure decreases can be realized.

Limitations

The use of hybrid defects as the ground truth instead of angiographically confirmed coronary artery occlusion, can be seen as a limitation, however, the ability to set the 'difficulty level' by a careful independent selection of defect size and severity can be helpful when confirming the optimization of processing strategies as described herein. Admittedly, a full clinical evaluation needs to follow this work.

NEW KNOWLEDGE GAINED

Based on our human-observer ROC study, the impact of reducing the acquired counts to 25% of the original and processing using an optimized OSEM strategy with AC, SC, and RC, is only a mild and statistically insignificant decrease in AUC compared to OSEM full-dose, but statistically significantly higher in AUC compared to FBP full-dose. Thus, a reduction of a factor of ~4 in count level, or quarter-dose, may be possible. Likewise, a reduction in dose can be combined with a reduction in acquisition time to realize half-time half-dose MPI.

CONCLUSIONS

We showed that cardiac perfusion SPECT/CT using optimized OSEM at a 25% dose level reconstructed with AC, SC, and RC, performed better in terms of AUC and specificity (true negative fraction) for hybrid perfusion-defect detection than optimized FBP at full-dose when read by human observers. The human observer AUC’s also outperforms the TPD clinical model observer AUC values for both OSEM at full dose and OSEM at a 25% dose level. However, human observers had lower AUC's compared to the clinical model observer when reading FBP full dose slices. The likely reason is the 'unfair' advantage the gender-specific databases provide when calculating the TPD scores. Finally, the performance of human observers in reading the optimized OSEM slices at a 25% dose level with AC, SC, and RC, compared well with the clinical model observer (TPD scores) performance for OSEM at full dose with AC, SC, and RC.

ABREVIATIONS

MPI

myocardial perfusion imaging

LNT

linear no threshold

QI

quality index

AC

attenuation compensation

SC

scatter compensation

RC

resolution compensation

AUC

area under the ROC curve

TPD

total perfusion deficit

OSEM

ordered-subset expectation-maximization

FBP

filtered backprojection