Abstract
Rationale: Ventilation–perfusion scintigraphy is highly sensitive for pulmonary embolism (PE), but its clinical usefulness is limited by its nondiagnostic rate. Objective analysis of single photon emission computed tomography (SPECT) three-dimensional scintigraphy may improve its diagnostic performance compared with subjective interpretation.
Objectives: To determine the diagnostic accuracy of objective SPECT analysis in PE.
Methods: We determined the ventilation/perfusion (V̇/Q̇) relationship using SPECT scintigraphy in a retrospective cohort of 73 patients. Measures of V̇/Q̇ heterogeneity (logSDQ, logSDV, logSDVQR), including a novel parameter, the weighted median V̇/Q̇ value, were calculated. Using receiver operating characteristic (ROC) analysis, each parameter's diagnostic accuracy was determined. The weighted median V̇/Q̇ value was then assessed prospectively in a second cohort of 50 patients.
Measurements and Main Results: In cohort 1, all parameters of V̇/Q̇ heterogeneity were higher in patients with PE (p < 0.002). The weighted median V̇/Q̇ had the highest area under the ROC curve (0.93; 95% confidence interval, 0.87–0.98). When applied to the prospective cohort, the area under the ROC curve was 0.87 (95% confidence interval, 0.75–0.99), with diagnostic cutoff values having negative and positive predictive values of 96 and 83%, respectively. In the retrospective and prospective cohorts, 82 and 73% of initially reported intermediate or low probability scans had diagnostic weighted median V̇/Q̇ values, with 90 and 100% accuracy, respectively.
Conclusions: Objective analysis of SPECT scintigraphy has a high diagnostic accuracy in patients with suspected PE. Objective analysis has the potential to reduce the number of nondiagnostic scan results, and may be useful for quantifying V̇/Q̇ mismatch in other pulmonary disorders.
No work has been done to date on the objective analysis of ventilation–perfusion single photon emission computed tomography (SPECT) scintigraphy to investigate ventilation/perfusion (V̇/Q̇) relationships in patients with potential pulmonary embolism.
Objective analysis of SPECT scintigraphy has a high diagnostic accuracy in patients with suspected pulmonary embolism.
Several authors have proposed that objective analysis, rather than subjective interpretation, may improve ventilation–perfusion scintigraphy's diagnostic performance (5–13). Attempts to analyze the ventilation/perfusion (V̇/Q̇) relationship using two-dimensional planar scans have been limited by high levels of image noise, difficulties in image registration, and poor spatial contrast (10, 14, 15). Single photon emission computed tomography (SPECT) provides three-dimensional images, which allow more accurate determination of each voxel's V̇/Q̇ value. Accordingly, SPECT has been shown to be more sensitive and specific than planar imaging in PE diagnosis clinically (16–20), and analysis of the SPECT-derived V̇/Q̇ profile has been shown to correlate well with the gold standard of V̇/Q̇ analysis—the Multiple Inert Gas Elimination Technique (MIGET) (21). SPECT ventilation and perfusion images have recently been used to produce three-dimensional quotient and subtraction parametric maps (19, 22). These have been used as visual aids to the subjective interpretation of scintigraphy in PE diagnosis (13); however, to date, no method has been explored that is entirely objective. We hypothesized that, because PE typically results in areas of lung with high V̇/Q̇ values, the SPECT-derived V̇/Q̇ distribution will be altered in patients suffering from this disease. Furthermore, objective parameters that quantify V̇/Q̇ heterogeneity are likely to be clinically useful in the diagnosis of PE.
In this study, we use SPECT scintigraphy to determine the V̇/Q̇ relationship of patients with suspected PE. In a retrospective population, we first determined the diagnostic values of several parameters of V̇/Q̇ heterogeneity, including the log10 of the standard deviation of ventilation (logSDV) and perfusion (logSDQ). In addition, based on the premise that PE results in the lung being composed of a number of distinct functional subpopulations, we evaluated a novel parameter of V̇/Q̇ heterogeneity, termed the “weighted median V̇/Q̇ value” and found it to be the most accurate parameter with respect to PE diagnosis. In a second prospective population of 50 patients, we measured the diagnostic accuracy of predetermined cutoff values for the weighted median V̇/Q̇ and the ability of objective analysis to reduce the number of nondiagnostic results. Some results of this study have been previously reported in abstract form (23, 24).
Two study populations were enrolled. Study population 1 is a retrospectively selected cohort, recruited from Royal North Shore Hospital, Sydney, Australia, from which the diagnostic performance of objective SPECT analysis was investigated, and cutoff values for the diagnosis of PE were established. Study population 2 is a prospectively collected cohort with suspected PE, recruited from John Hunter Hospital, Newcastle, Australia. This is a clinical population in which the final diagnosis was based on multiple investigations and on which the diagnostic performance of the cutoff values were validated. This study was approved by the human ethics research committees of the Northern Sydney and Central Coast Area Health Service (protocol 0512–232M), and the Hunter New England Area Health Service (protocol 03/10/15/3.08).
All patients presenting to the Department of Nuclear Medicine who underwent both SPECT scintigraphy and CTPA for the investigation of potential PE between September 2004 and January 2006 were retrospectively identified, and their medical records reviewed. Patients with a documented past medical history of PE were excluded. Due to the potential influence of concurrent anticoagulation, patients with a final clinical diagnosis of PE were excluded if the scintigraphy and CTPA imaging were more than 72 hours apart (25). Final clinical diagnosis was established by a consensus panel of three experienced physicians, who retrospectively reviewed all clinical and imaging data. This included compression ultrasonography and digital subtraction angiography, where available.
Consecutive patients presenting for investigation of suspected PE were identified for recruitment. Patients were excluded if they had a past medical history of PE, renal dysfunction, or significant contrast hypersensitivity. Patients were also excluded if they were unable to comply with the study protocol, or if imaging was unable to be completed within 24 hours of presentation. Consenting patients underwent both planar and SPECT ventilation–perfusion scintigraphy, and CTPA/CTV. Patients younger than 50 years were excluded from recruitment due to the study protocol requiring patients to undergo both scintigraphy and CTPA to best establish a confident diagnosis. During the investigation period, patients were treated with anticoagulants at the discretion of the treating physician. Patients without a clinical diagnosis of acute PE were contacted at 3 months by telephone to determine if there had been any episodes of possible recurrence. Final clinical diagnosis was determined by a consensus panel of three experienced physicians who had access to all clinical details, including the results of planar scintigraphy, CTPA, and CTV.
Both cohorts initially underwent supine inhalation of technetium Tc 99m (99mTc) Technegas (Vita Medical, Sydney, Australia) to ensure 40 to 50 MBq activity within the lungs. If patients required supplemental oxygen, Technegas inhalation was performed during a brief pause (< 10 s) in the oxygen delivery. SPECT acquisition was performed using 3° steps through a 360° acquisition. Without moving the patient, 160 to 220 MBq of 99mTc macroaggregated albumin (MAA) Pulmolite (Dupont, Wilmington, DE) was injected intravenously while the patient was supine, and perfusion SPECT data were acquired in a similar manner. Total acquisition time was approximately 20 minutes on a dual-headed camera, and 14 minutes on a triple-headed camera. All SPECT data were reconstructed to 128 slices using Ordered Subset Expectation Maximization iterative reconstruction (8 iterations, 4 subsets), and smoothed with a postreconstruction three-dimensional Butterworth filter. The ventilation and perfusion images were registered to each other using an automated procedure (HERMES multimodality program; Nuclear Diagnostics, Stockholm, Sweden), which uses a mutual information algorithm (26). For the purposes of clinical reporting, patients either underwent additional traditional six-view planar acquisition or had planar images regenerated from SPECT data using a reprojection method (27). All scans were reported by an experienced nuclear medicine physician, who had the results of recent chest radiographs available. Reports were issued according to standard modified PIOPED reporting criteria (28).
Patients underwent CTPA on either a 4- or 16-slice multidetector spiral scanner, depending on the availability at each institution. Data were acquired using an automated CTPA protocol (120 kVp, 220–240 mAs, slice thickness of 2.0 or 2.5 mm). Scans were performed using an automatically triggered injection of 80 to 120 ml of intravenous contrast (Ultravist 370; Schering, Berlin, Germany). Image data were reconstructed in a 512 × 512 matrix and stored in Digital Imaging and Communication in Medicine (DICOM) format on a dedicated radiology server. Scans were reviewed in a darkened environment by an experienced clinical radiologist on an interactive workstation allowing visual tracking of the pulmonary arteries over contiguous slices. If necessary, coronal and sagittal reformatting was performed for review.
Using datasets coregistered with the automated procedure described above, perfusion data were corrected for remaining ventilation activity by image subtraction after decay correction. Given that the total ventilation photons counted (“counts”) are typically three to four times lower than the total perfusion counts, when combined with radioactive decay, this subtraction has negligible effect on image noise. Any deposition of tracer within the oropharynx, trachea, major bronchi, and stomach was manually masked by an experienced operator blinded to the final clinical diagnosis. Using the Interactive Data Language (IDL) programming environment (RSI, Boulder, CO), a threshold value of 10% was applied to the composite of the ventilation and perfusion datasets to define the lung boundary (17, 19, 29). The total number of ventilation counts was then normalized to the total corrected perfusion counts. A V̇/Q̇ dataset was subsequently generated as previously described by dividing the normalized ventilation count by the perfusion count for each voxel (19, 22) (Figure 1).
Figure 1. Generation of a V̇/Q̇ dataset in a patient with a high probability of pulmonary embolism. The top row demonstrates tomographic sections of a ventilation single photon emission computed tomography scan in coronal, transverse, and sagittal views. The middle row shows the corresponding ventilation corrected perfusion images, and the third row shows the V̇/Q̇ images. Images are displayed using a color table in which black represents low V̇/Q̇ values and white high V̇/Q̇ values.
[More] [Minimize]The V̇/Q̇ data were log10 transformed, and separated into 101 equal intervals, ranging from values of −3 (= log10 0.001) to 2 (= log10 100) with a width of 0.05. These values were chosen to correspond with those used for the MIGET (30). All voxels with a value of more than 100 were assigned a value of 100, whereas those with a value below 0.001 were assigned a value of 0.001. The frequency distribution of the V̇/Q̇ dataset was then constructed, together with the relative distribution of total ventilation and total perfusion to each V̇/Q̇ interval.
We determined several parameters to characterize heterogeneity of the V̇/Q̇ relationship, and investigated the performance of these in study population 1. First, we derived the logSD of the perfusion (logSDQ), ventilation (logSDV), and the V̇/Q̇ distribution (logSDVQR) curves. We also calculated the fraction of lung volume with a V̇/Q̇ value above the threshold values of 2.5 (Vol%>2.5), 5 (Vol%>5), and 10 (Vol%>10). In addition, we hypothesized that, because PE results in impaired perfusion to one or more populations of functional lung units, these populations should have a V̇/Q̇ relationship that is different to the surrounding lung. Furthermore, based on MIGET analysis (30), each of these distinct functional populations should have a log-normal V̇/Q̇ distribution. Therefore, by iteratively fitting multiple log-normal curves, as described in the online supplement, we derived a novel parameter termed the weighted median V̇/Q̇ value. This parameter potentially more accurately describes deviation of the V̇/Q̇ distribution from normal, by handling each functional subpopulation individually.
Statistical analyses were performed using the Analyze-it statistics program (Analyze-it Software Ltd., Leeds, UK). The Mann-Whitney U test with correction for ties was performed to compare nonparametric independent samples, whereas the chi-square test was used to test for differences in proportions between study populations (31). Receiver operating characteristic (ROC) analysis was used to determine the sensitivity and specificity of the V̇/Q̇ distribution analyses (32). From this, values that were closest to being 95% sensitive and 95% specific were chosen as diagnostic cutoffs. Comparison was made between the area under the ROC curves using the method of Hanley and McNeil (33). The likelihood of PE in the derived categories was expressed as a likelihood ratio (34). Spearman rank correlation was used to assess any relationship between nonparametric continuous variables (35).
Seventy-three patients meeting inclusion criteria were identified. Their demographic data are presented in Table 1. Fifty-five patients had scintigraphy as the initial investigation, whereas 18 patients underwent CTPA as the initial investigation. Overall, the incidence of PE in this cohort was 26%. The results of the clinical interpretation of the ventilation–perfusion scans, and corresponding rates of PE diagnosis, are shown in Table 4.
Study Population 1 | Study Population 2 | p Value | |
|---|---|---|---|
| No. | 73 | 50 | |
| Age, yr (mean ± SD) | 61 ± 19 | 66 ± 13 | 0.15 |
| Male/female, % | 37/63 | 64/36 | 0.006 |
| Nonsmoker, % | 46 | 58 | 0.12 |
| Smoker, % | 54 | 42 | 0.12 |
| Current | 14 | 10 | 0.74 |
| Ex | 40 | 32 | 0.49 |
| Lung disease, % | 27 | 38 | 0.23 |
| COPD | 5 | 14 | 0.19 |
| Asthma | 14 | 18 | 0.69 |
| Other | 12 | 12 | 0.82 |
| Heart disease, % | 19 | 44 | 0.006 |
| Active malignancy, % | 14 | 16 | 0.92 |
| Outpatient/inpatient referral, % | 63/37 | 58/42 | 0.71 |
All V̇/Q̇ distributions were satisfactorily modeled using a median of 4 log-normal curves (range, 2–8), yielding a median R2 value for the iterative modeling of 0.9997 (range, 0.9858–1.0000). Examples of V̇/Q̇ distributions from patients with and without a final diagnosis of PE are shown in Figure 2. There was a significant difference in all quantitative parameters between patients who suffered PE and those who had not (Table 2). This difference was greatest for the weighted median V̇/Q̇ value (Mann-Whitney U statistic = 1,029, p < 0.0001). There were no differences in the R2 value (Mann-Whitney U statistic = 622, p = 0.17) or the number of log-normal curves used to fit the V̇/Q̇ distribution (Mann-Whitney U statistic = 461, p = 0.50) for the two clinical categories.
Figure 2. Single photon emission computed tomography (SPECT)–derived V̇/Q̇ frequency distributions (A and C) and plots of ventilation and perfusion against V̇/Q̇ (B and D) in two patients. A and B show the same patient without pulmonary embolism; C and D show the same patient with pulmonary embolism. Note that, in B, the perfusion curve is lying directly on top of the ventilation curve.
[More] [Minimize]Clinical Diagnosis | ||||
|---|---|---|---|---|
| Parameter | Pulmonary Embolism | No Pulmonary Embolism | p Value | |
| LogSDV | 0.25 (0.23–0.31) | 0.18 (0.15–0.21) | < 0.0001 | |
| LogSDQ | 0.18 (0.18–0.23) | 0.17 (0.14–0.20) | < 0.002 | |
| LogSDVQR | 0.27 (0.25–0.31) | 0.21 (0.17–0.24) | < 0.0001 | |
| Vol%>10 | 0.95 (0.38–2.5) | 0.077 (0.00–0.25) | < 0.0001 | |
| Vol%>5 | 2.1 (1.1–4.4) | 0.34 (0.04–0.65) | < 0.0001 | |
| Vol%>2.5 | 7.6 (5.7–14.0) | 2.7 (1.5–4.6) | < 0.0001 | |
| Weighted median V̇/Q̇ | 2.0 (1.6–5.9) | 1.2 (1.1–1.4) | < 0.0001 | |
The results of ROC analyses for each of the quantitative methods (logSDQ, logSDV, logSDVQR, Vol%>2.5, Vol%>5, Vol%>10, and weighted median V̇/Q̇) are shown in Table 3. The weighted median V̇/Q̇ had the highest area under the curve (AUC) of 0.93 (95% confidence interval [CI], 0.87–0.98) (Figure 3). This was significantly higher than the AUC for logSDQ (p = 0.001), logSDVQR (p = 0.03), and Vol%>10 (p = 0.04), but not for logSDV (p = 0.07), Vol%>5 (p = 0.11), or Vol%>2.5 (p = 0.44).
Figure 3. The receiver operating characteristic curves for objective analysis of the single photon emission computed tomography (SPECT)–derived weighted median V̇/Q̇ value in study populations 1 and 2. Results are expressed as area under the curve (AUC) (95% confidence interval).
[More] [Minimize]Parameter | AUC (95% CI) |
|---|---|
| LogSDQ | 0.71 (0.59–0.83) |
| LogSDVQR | 0.82 (0.72–0.92) |
| LogSDV | 0.84 (0.75–0.94) |
| Vol%>10 | 0.81 (0.68–0.95) |
| Vol%>5 | 0.86 (0.75–0.96) |
| Vol%>2.5 | 0.90 (0.83–0.97) |
| Weighted median V̇/Q̇ | 0.93 (0.87–0.98) |
From the ROC analyses, a cutoff weighted median V̇/Q̇ value of 1.35 was 95% sensitive with a negative predictive value (NPV) of 98% (likelihood ratio = 0.07), whereas a value of 1.83 was 95% specific, with a positive predictive value (PPV) of 81% (likelihood ratio = 12.3). Given this, we considered a weighted median V̇/Q̇ value of below 1.35 to indicate that PE was unlikely, whereas a value above 1.83 indicated that PE was likely. A weighted median V̇/Q̇ value between 1.35 and 1.83 was considered to be nondiagnostic (likelihood ratio = 1.42). Of the 73 patients in the initial cohort, 42 (58%) had a weighted median V̇/Q̇ value below 1.35, whereas 16 (22%) had a weighted median V̇/Q̇ value above 1.83 (Table 3). Fifteen (21%) patients had a weighted median V̇/Q̇ value between 1.35 and 1.83. Of the original 49 patients who were reported to have either intermediate or low probability scans, 40 (82%) had a weighted median V̇/Q̇ value within the diagnostic range (35 of 40 had values less than 1.35, and 5 of 40 had values greater than 1.83). When correlated with the final clinical diagnosis, the overall accuracy for diagnosing PE in this subgroup of patients with diagnostic weighted median V̇/Q̇ values was 90%.
Given that study population 2 consisted only of patients older than 50 years, and that this may influence any predetermined cutoff values, we investigated the relationship between age and the weighted median V̇/Q̇ value. In patients without PE as a final clinical diagnosis, we demonstrated a correlation between age and the weighted median V̇/Q̇ value (r = 0.30, p = 0.03), and concluded that an increased population age would alter the cutoff values. We therefore determined relevant cutoff values using ROC analysis on the subgroup of study population 1 who were older than 50 years (51 patients). This yielded an AUC of 0.94 (95% CI, 0.87–1.00) for the weighted median V̇/Q̇ value, with a value of 1.41 being 93% sensitive (NPV = 96%, likelihood ratio = 0.09), and a value of 2.17 being 94% specific (PPV = 89%, likelihood ratio = 19). Fourteen patients (27%) had a weighted median V̇/Q̇ in the nondiagnostic range (likelihood ratio = 1.8).
Fifty patients meeting inclusion criteria were enrolled. Demographics of this cohort are presented in Table 1. Overall, the incidence of PE in this cohort was 28%. Results of the clinical interpretation of the ventilation–perfusion scans, and the corresponding rates of PE in each interpretive category, are shown in Table 4.
Initial Scan Interpretation | Number (%) | Number with PE (%) | Number with a Diagnostic Weighted Median V̇/Q̇ (%) | Accuracy of Objective Analysis in Patients with a Diagnostic Weighted Median V̇/Q̇ Value (%) |
|---|---|---|---|---|
| Study population 1 | ||||
| High | 19 (26) | 17 (89) | 15 (79) | 93 |
| Intermediate | 36 (49) | 2 (5) | 31 (86) | 90 |
| Low | 13 (18) | 0 (0) | 8 (62) | 100 |
| Normal | 5 (7) | 0 (0) | 4 (80) | 100 |
| All categories | 73 | 19 (26) | 58 (79) | 93 |
| Study population 2 | ||||
| High | 8 (16) | 8 (100) | 7 (88) | 100 |
| Intermediate | 3 (6) | 2 (66) | 2 (66) | 100 |
| Low | 23 (46) | 3 (13) | 17 (74) | 100 |
| Normal | 16 (32) | 1 (6) | 9 (56) | 66 |
| All categories | 50 | 14 (28) | 35 (70) | 91 |
We assessed the diagnostic accuracy of the weighted median V̇/Q̇ value, given that it had the greatest diagnostic value of all V̇/Q̇ parameters in study population 1. All V̇/Q̇ distributions were again satisfactorily modeled using a median of 4 log-normal curves (range, 2–9), with a median R2 value for the iterative modeling of 0.9997 (range, 0.9892–1.0000). Using this parameter, there was a significant difference between patients with and without PE (Mann-Whitney U statistic = 432, p < 0.001). Using cutoff values determined from study population 1 for the subgroup of patients older than 50 years, ROC analysis demonstrated an AUC of 0.87 (95% CI, 0.75–0.99). Thirty-five of the 50 scans were classified into the PE-likely or -unlikely categories by weighted median V̇/Q̇ criteria, with 15 patients (30%) falling in the nondiagnostic range (likelihood ratio = 0.64). The predetermined cutoff values of 1.41 and 2.17 had a likelihood ratio of 0.12 (NPV, 96%,) and 13 (PPV, 83%), respectively. In patients who had a weighted median V̇/Q̇ value in the diagnostic range, the overall diagnostic accuracy was 91%. Within study population 2, 26 patients (52%) had scans initially reported as intermediate or low probability. Of these, 19 patients (73%) had weighted median V̇/Q̇ values in the diagnostic ranges, whereas 7 (27%) remained in the nondiagnostic range. When correlated with the final clinical diagnosis, the overall accuracy for diagnosing PE in this subgroup of patients with a diagnostic weighted median V̇/Q̇ value was 100% (see Table 4).
In this study, we describe the use of SPECT ventilation–perfusion scintigraphy to derive objective parameters regarding the V̇/Q̇ relationship. Although there are many potential applications for this technique, here we have demonstrated its performance in the frequently difficult area of PE diagnosis. By determining a novel continuous parameter, the weighted median V̇/Q̇, we have shown it has high diagnostic accuracy, with an AUC in the retrospective and prospective populations of 0.93 and 0.87, respectively. This compares favorably to previously published data, in which the AUC for CTPA was 0.94 (36). In addition, by deriving sensitive and specific cutoff values for the diagnosis and exclusion of PE, we have demonstrated the potential of this technique to accurately reclassify the majority of scans that have a nondiagnostic clinical interpretation.
Given that our analysis used ROC methodology to compare the results of objective V̇/Q̇ analysis with clinical outcome, it was critically important to ensure that the final diagnosis was accurate. The gold standard for PE has long been considered to be pulmonary angiography. However, given that is not commonly performed in a clinical setting, and is known to have flaws as a diagnostic test (37), recent studies have favored a composite gold standard to determine the presence or absence of PE (3). We chose a similar approach, with all patients undergoing both scintigraphy and CTPA. From these, together with appropriate clinical information, the results of other relevant investigations, and clinical follow-up in the case of study population 2, a consensus panel of experienced physicians determined the final clinical diagnosis. Although this approach ensured a minimum of two imaging investigations for enrolled patients, it resulted in study population 2 being limited to patients older than 50 years at the request of the radiation safety committee. On the presumption that age may affect the weighted median V̇/Q̇ value, we examined this in patients from study population 1 without PE. Subsequently, we demonstrated a significant correlation between these two variables (r = 0.30), and adjusted the relevant diagnostic cutoff values accordingly. However, despite this, a small difference in the results of ROC analysis between the two populations was present. Some of this discrepancy is potentially explained by other differences between the populations, such as the greater prevalence of coexistent cardiac disease in study population 2.
Although the prevalence of PE in the two study populations was similar, differences in the reporting patterns according to probability criteria were evident. Forty-nine percent of scans in the retrospective cohort were interpreted as intermediate compared with only 6% in the prospective cohort. This high rate in study population 1 is likely to be partly explained by the selection criteria, with patients who had an initial intermediate probability scintigraphy scan being much more likely to undergo further imaging with CTPA because of clinical indications. The low rate of intermediate scans in study population 2 may be due to local differences in reporting patterns, with this cohort having a correspondingly higher rate of low probability scans (46% compared with 18% in study population 1). Importantly, these differences highlight the impact of interobserver variability on subjective reporting, something that an objective analysis technique, such as the method we describe in this article, could go a long way toward reducing or even eliminating.
Although we investigated several different parameters of V̇/Q̇ heterogeneity, the weighted median V̇/Q̇ value was the most diagnostic, as determined by ROC analysis. The measurement of this parameter is based on the hypothesis that the overall V̇/Q̇ distribution is made up of subpopulations, each with different functional characteristics to the rest of the lung. This is particularly relevant to the diagnosis of PE, where vascular occlusion results in one or more regions of the lung having higher V̇/Q̇ values than the surrounding unaffected lung. This typically results in a V̇/Q̇ distribution that is skewed to the high V̇/Q̇ end, which, when sufficiently severe, may be multimodal in nature. In contrast, we have observed that parenchymal lung disease typically broadens the traditional unimodal V̇/Q̇ relationship, without causing significant skewness. Consequently, we speculate that by attempting to model the underlying physiology, the weighted median V̇/Q̇ value is better able to account for both the skewness and dispersion of the V̇/Q̇ distribution caused by PE. In keeping with this, in patients without PE, we were able to demonstrate no significant difference of the weighted median V̇/Q̇ value in patients with coexistent lung disease compared with those without (Mann-Whitney U statistic = 976, p = 0.43).
Although the other parameters of V̇/Q̇ heterogeneity had a lower diagnostic accuracy, they still offer potentially useful information. The logSDQ and logSDV have been extensively used as objective parameters in studies using the MIGET (31). Because they represent measurements of standard deviation, they are more suited to the analysis of normally distributed rather than skewed data. This is the likely reason for their lower diagnostic accuracy in our cohort. Nonetheless, the logSDQ is particularly useful for measuring V̇/Q̇ heterogeneity in areas of low V̇/Q̇ (< 1.0), whereas the logSDV is a measure of V̇/Q̇ heterogeneity in areas of high V̇/Q̇ (> 1.0). The logSDVQR, on the other hand, provides a general measure of V̇/Q̇ heterogeneity. Given that the major disturbance in PE is one of high rather than low V̇/Q̇, this is the likely explanation for why the logSDV had a greater accuracy than logSDQ in our cohort.
In addition to demonstrating a high overall diagnostic accuracy, we have also shown the ability of objective analysis to potentially reduce the number of nondiagnostic scintigraphy results. If intermediate and low probability scans are considered to be nondiagnostic, then 67 and 52% of study populations 1 and 2, respectively, would require further diagnostic testing. This clearly has cost implications due to the additional length of hospitalization and number of investigations these patients would require. The weighted median V̇/Q̇ value performed favorably in these subgroups of patients when it was in the diagnostic range, reclassifying 80 and 73% of study populations 1 and 2 with 92 and 100% accuracy, respectively.
Ventilation–perfusion SPECT scintigraphy is based on the fundamental principle that the number of photons originating from an area of lung is proportional to the relative distribution of the agent being imaged. Consequently, perfusion images provide information on the relative topographical distribution of cardiac output, whereas ventilation images provide similar information on alveolar ventilation. Because the absolute values for the cardiac output and alveolar ventilation were not known, we normalized the data such that the total number of ventilation counts equaled the total perfusion counts. In doing this, we made the physiological assumption that alveolar ventilation was equal to the cardiac output. However, in the setting of PE, this is likely to be incorrect, and could potentially influence SPECT-derived parameters of V̇/Q̇ heterogeneity. Consequently, we are currently exploring the use of exhaled gas analysis with acetylene uptake to allow adjustment of the measurements for cardiac output and alveolar ventilation. Accurate measurements in critically ill patients, however, may be difficult to obtain.
One potential source of error in calculating the V̇/Q̇ distribution is the process of spatially registering the ventilation and perfusion images. This was performed using automated software and the mutual information algorithm (26). This technique involves iterative minimization of the differences between two datasets, and resulted in all datasets being satisfactorily coregistered by visual inspection. We did not use radioactive fiducial surface markers because changes in spinal posture and breathing patterns can make these techniques insensitive to significant errors of registration (38). The previously described technique of simultaneous dual isotope acquisition with indium In 113m (113mIn) MAA and 99mTc Technegas could be used to eliminate this source of error (39). By contemporaneously acquiring both ventilation and perfusion, and then separating the signals based on the different isotope γ-ray energies, the images would be inherently registered without the use of software. However, the use of 113mIn would make this technique significantly more costly and it is currently unavailable for such use in Australia.
Subsequent to the analysis of the PIOPED data, PE has been known to cause matched perfusion defects (40). However, the true rate of occurrence of these defects is debated. It has been suggested that the PIOPED trial overestimated the rate of truly matched defects because xenon Xe 133 (133Xe) ventilation imaging lacks appropriate spatial resolution (41). In our subjects, although some defects were believed subjectively to be matched, when analyzed objectively this was often not the case. Nonetheless, our technique, which relies on detecting V̇/Q̇ mismatch, will have limited ability to detect PE where predominantly matched defects are present. This is illustrated by the observation that patients who had PE with matched defects tended to have a weighted median V̇/Q̇ value lower than those without. Consequently, although we have demonstrated that quantitative parameters of V̇/Q̇ mismatch are useful in PE diagnosis, the need for concurrent clinical interpretation is likely to remain.
We used Technegas, an ultrafine carbon particle with a median diameter of less than 200 nm, labeled with 99mTc (42, 43), to image ventilation. It has been shown to have a similar distribution to an inert gas (44, 45); however, when it impacts with alveolar structures, it shows no appreciable elimination or absorption. This makes it an extremely useful agent for ventilation SPECT imaging (45). In patients with airflow limitation, impaction of Technegas particles in central airways sometimes occurs, resulting in “hot spots” of activity. When compared with perfusion, these can result in voxels with low V̇/Q̇ values, consequently falsely skewing the V̇/Q̇ distribution. We therefore removed these by manually masking areas of central deposition, which we did on both the ventilation and perfusion datasets concurrently. Future useful improvements to deal with this significant imaging problem include automated masking of hot spots and control of the inspiratory technique used for Technegas delivery.
Although we chose Technegas, other radiotracers have been used for imaging ventilation with SPECT. Krypton Kr 81m is a radioactive gas with a very short half-life (13 s). To perform SPECT acquisition, it requires continuous administration, and is therefore difficult if patients require concomitant oxygen. Although its being a true gas and having a short half-life make it suited to imaging regional ventilation, it is prone to error in patients with an increased minute ventilation (46). In addition, its need for a specialized rubidium (81Rb)–krypton generator, and the limited availability and cost of its parent isotope (81Rb), significantly limit its availability, and result in substantial ongoing costs. SPECT acquisition of 133Xe has also been described (47). However, as a result of its longer half-life, it is prone to error from inhomogeneous clearance into the pulmonary circulation, and recirculation (48). As previously mentioned, its low energy (81 keV) also results in poorer spatial resolution. Finally, nebulized 99mTc–diethylene triamine pentaacetic acid (DTPA) can be used to produce an inexpensive aerosol that has been used to image ventilation. However, because of the relatively large mass median aerodynamic diameter of the droplets (2–5 μm), it is particularly prone to central airway deposition and clumping. It is also cleared into the pulmonary circulation, or via the mucociliary apparatus, both of which can occur during the SPECT acquisition, resulting in image distortion.
Finally, SPECT imaging and objective analysis, as described here, can be easily and cheaply implemented in most nuclear medicine imaging facilities with SPECT capability. Once acquired, images can then be processed either on site or remotely. After coregistration and manual masking, the computations can be automated, requiring minimal user intervention. Our current algorithms allow processing and analysis within 5 to 10 minutes per patient. It should also be noted that improvements in SPECT scanning are likely to increase the clinically utility of objective scintigraphy analysis further. In particular, developments in the processing of SPECT data promise to improve the spatial resolution of this technique, whereas the increased availability of combined SPECT– computed tomography will allow accurate definition of the edge of the lung, and colocalization of ventilation and perfusion abnormalities with lung anatomy and parenchymal abnormalities. This will allow V̇/Q̇ analysis to be performed at a lobar or segmental level, rather than for the whole lung, and will offer anatomical rather than statistical separation of lung populations. This is likely to be particularly useful for understanding the regional physiology of patients with diseases such as emphysema and asthma, where the ability to match ventilation to perfusion can vary regionally.
Objective analysis of SPECT ventilation–perfusion scintigraphy provides a useful tool for investigating regional V̇/Q̇ relationships. It is clinically relevant in the diagnosis of PE, and may help reduce the number of nondiagnostic scintigraphy results. It provides objective measures of V̇/Q̇ mismatch, is easy to implement, and is likely to have applications in the physiological investigation of other pulmonary diseases.
The authors thank Laurie S. Von Dauber for help with programming Origin software, and Catherine Walsh for help in collating data. They also thank all the staff in the Departments of Respiratory Medicine, Nuclear Medicine, and Radiology at Royal North Shore Hospital, Sydney, and John Hunter Hospital, Newcastle, for their assistance.
| 1. | Wagner HN, Sabiston DC, McAfee JG, Tow D, Stern HS. Diagnosis of massive pulmonary embolism in man by radioisotope scanning. N Engl J Med 1964;271:377–384. |
| 2. | The PIOPED Investigators. Value of the ventilation/perfusion scan in acute pulmonary embolism: results of the prospective investigation of pulmonary embolism diagnosis (PIOPED). JAMA 1990;263:2753–2759. |
| 3. | Stein PD, Fowler SE, Goodman LR, Gottschalk A, Hales CA, Hull RD, Leeper KV, Jr, Popovich J, Jr, Quinn DA, Sos TA, et al. Multidetector computed tomography for acute pulmonary embolism. N Engl J Med 2006;354:2317–2327. |
| 4. | Parker MS, Hui FK, Camacho MA, Chung JK, Broga DW, Sethi NN. Female breast radiation exposure during CT pulmonary angiography. AJR Am J Roentgenol 2005;185:1228–1233. |
| 5. | Kramer EL, Sanger JJ. 81mKr gas and 99mTc-MAA V̇/Q̇ ratio images for detection of V̇/Q̇ mismatches. Eur J Nucl Med 1984;9:345–350. |
| 6. | Sando Y. Assessment of pulmonary ventilation/perfusion ratio by dual radionuclides SPECT using 81mKr gas and 99mTc-MAA [in Japanese]. Kaku Igaku 1994;31:1209–1218. |
| 7. | Sando Y, Inoue T, Nagai R, Endo K. Ventilation/perfusion ratios and simultaneous dual-radionuclide single-photon emission tomography with krypton-81m and technetium-99m macroaggregated albumin. Eur J Nucl Med 1997;24:1237–1244. |
| 8. | Arnold JE, Wilson BC. Computer processing of perfusion, ventilation, and V̇/Q̇ images to highlight pulmonary embolism. Eur J Nucl Med 1981;6:309–315. |
| 9. | Nakata Y, Narabayashi I, Sueyoshi K, Matsui R, Namba R, Tabuchi K. Evaluation of the ventilation-perfusion ratio in lung diseases by simultaneous anterior and posterior image acquisition. Ann Nucl Med 1994;8:269–276. |
| 10. | Itti E, Nguyen S, Robin F, Desarnaud S, Rosso J, Harf A, Meignan M. Distribution of ventilation/perfusion ratios in pulmonary embolism: an adjunct to the interpretation of ventilation/perfusion lung scans. J Nucl Med 2002;43:1596–1602. |
| 11. | Meignan M, Simonneau G, Oliveira L, Harf A, Cinotti L, Cavellier JF, Duroux P, Ansquer JC, Galle P. Computation of ventilation-perfusion ratio with Kr-81m in pulmonary embolism. J Nucl Med 1984;25:149–155. |
| 12. | Teertstra HJ, Janssen PJ, Posch C, Verdegaal WP. The use of a computer in ventilation-perfusion scintigraphy with Tc-99m macroaggregates and Kr-81m, to obtain a physiological model. Eur J Nucl Med 1987;13:24–27. |
| 13. | Reinartz P, Kaiser H-J, Wildberger JE, Gordji C, Nowak B, Buell U. SPECT imaging in the diagnosis of pulmonary embolism: automated detection of match and mismatch defects by means of image-processing techniques. J Nucl Med 2006;47:968–973. |
| 14. | Holst H, Åström K, Järund A, Palmer J, Heyden A, Kahl F, Trägil K, Evander E, Sparr G, Edenbrandt L. Automated interpretation of ventilation-perfusion lung scintigrams for the diagnosis of pulmonary embolism using artificial neural networks. Eur J Nucl Med 2000;27:400–406. |
| 15. | Frigyesi A. An automated method for the detection of pulmonary embolism in V̇/Q̇-scans. Med Image Anal 2003;7:341–349. |
| 16. | Reinartz P, Wildberger JE, Schaefer W, Nowak B, Mahnken AH, Buell U. Tomographic imaging in the diagnosis of pulmonary embolism: a comparison between V̇/Q̇ lung scintigraphy in SPECT technique and multislice spiral CT. J Nucl Med 2004;45:1501–1508. |
| 17. | Bajc M, Olsson CG, Olsson B, Palmer J, Jonson B. Diagnostic evaluation of planar and tomographic ventilation/perfusion lung images in patients with suspected pulmonary embolism. Clin Physiol Funct Imaging 2004;24:249–256. |
| 18. | Collart JP, Roelants V, Vanpee D, Lacrosse M, Trigaux JP, Delaunois L, Gillet JB, De Coster P, Vander Borght T. Is a lung perfusion scan obtained by using single photon emission computed tomography able to improve the radionuclide diagnosis of pulmonary embolism? Nucl Med Commun 2002;23:1107–1113. |
| 19. | Palmer J, Bitzen U, Jonson B, Bajc M. Comprehensive ventilation/perfusion SPECT. J Nucl Med 2001;42:1288–1294. |
| 20. | Magnussen JS, Chicco P, Palmer AW, Bush V, Mackey DW, Storey G, Magee M, Bautovich G, Van der Wall H. Single-photon emission tomography of a computerised model of pulmonary embolism. Eur J Nucl Med 1999;26:1430–1438. |
| 21. | Petersson J, Sánchez-Crespo A, Rohdin M, Montmerle S, Nyrén S, Jacobsson H, Larsson SA, Lindahl SG, Linnarsson D, Glenny RW, et al. Physiological evaluation of a new quantitative SPECT method measuring regional ventilation and perfusion. J Appl Physiol 2004;96:1127–1136. |
| 22. | Bajc M, Bitzén U, Olsson B, Perez de Sá V, Palmer J, Jonson B. Lung ventilation/perfusion SPECT in the artificially embolized pig. J Nucl Med 2002;43:640–647. |
| 23. | Harris BE, Bailey D, Roach P, King GG. Objective analysis of SPECT scintigraphy in pulmonary embolism using novel analysis of the V̇/Q̇ distribution [abstract]. Proc Am Thorac Soc 2006;3:A58. |
| 24. | Harris B, Bailey D, Roach P, Bailey E, King G. The objective use of SPECT scintigraphy to quantitate V̇/Q̇ relationships. J Nucl Med 2006;47(Suppl):357P. |
| 25. | Langer JE, Velchik MG. Rapid resolution of massive pulmonary embolism due to streptokinase therapy: documented by ventilation/perfusion imaging. Clin Nucl Med 1988;13:874–877. |
| 26. | Studholme C, Hill D, Hawkes DJ. Automated three-dimensional registration of magnetic resonance and positron emission tomography brain images by multiresolution optimization of voxel similarity measures. Med Phys 1997;24:25–35. |
| 27. | Bailey D, Schembri G, Cooper R, Bailey E, Roach P. Reprojection of reconstructed V̇/Q̇ SPECT scans to provide high count planar images. J Nucl Med 2005;46(Suppl):337P. |
| 28. | Gottschalk A, Sostman HD, Coleman RE, Juni JE, Thrall J, McKusick KA, Froelich JW, Alavi A. Ventilation-perfusion scintigraphy in the PIOPED study. Part II: evaluation of the scintigraphic criteria and interpretations. J Nucl Med 1993;34:1119–1126. |
| 29. | Palla A, Bellina CR, Marini C, Pazzagli M, Giuntini C. A non-invasive, quantitative method to demonstrate the early effects of therapy in acute pulmonary embolism. Eur J Nucl Med Mol Imaging 2001;28:1605–1609. |
| 30. | Roca J, Wagner P. Contribution of multiple inert gas elimination technique to pulmonary medicine. 1. Principles and information content of the multiple inert gas elimination technique. Thorax 1994;49:815–824. |
| 31. | Siegel S, Castellan N. Non-parametric statistics for the behavioral sciences. New York: McGraw-Hill; 1988. |
| 32. | Zweig MH, Campbell G. Receiver-operating characteristic (ROC) plots: a fundamental evaluation tool in clinical medicine. Clin Chem 1993;39:561–577. |
| 33. | Hanley JA, McNeil BJ. A method of comparing the areas under receiver operating characteristic curves derived from the same cases. Radiology 1983;148:839–843. |
| 34. | Deeks JJ, Altman DG. Diagnostic tests 4: likelihood ratios. BMJ 2004;329:168–169. |
| 35. | Conover W. Practical nonparametric statistics. New York: Wiley; 1980. |
| 36. | Safriel Y, Zinn H. CT pulmonary angiography in the detection of pulmonary emboli: a meta-analysis of sensitivities and specificities. Clin Imaging 2002;26:101–105. |
| 37. | Baile EM, King GG, Müller NL, D'Yachkova Y, Coche EE, Paré PD, Mayo JR. Spiral computed tomography is comparable to angiography for the diagnosis of pulmonary embolism. Am J Respir Crit Care Med 2000;161:1010–1015. |
| 38. | Gutman F, Hangard G, Gardin I, Varmenot N, Pattyn J, Clement JF, Dubray B, Vera P. Evaluation of a rigid registration method of lung perfusion SPECT and thoracic CT. AJR Am J Roentgenol 2005;185:1516–1524. |
| 39. | Sánchez-Crespo A, Petersson J, Nyrén S, Mure M, Glenny RW, Thorell JO, Jacobsson H, Lindahl SG, Larsson SA. A novel quantitative dual-isotope method for simultaneous ventilation and perfusion lung SPET. Eur J Nucl Med Mol Imaging 2002;29:863–875. |
| 40. | Gottschalk A, Stein P, Henry J, Relyea B. Matched ventilation, perfusion and chest radiographic abnormalities in acute pulmonary embolism. J Nucl Med 1996;37:1636–1638. |
| 41. | Schümichen C. V̇/Q̇-scanning/SPECT for the diagnosis of pulmonary embolism. Respiration (Herrlisheim) 2003;70:329–342. |
| 42. | Lemb M, Oei TH, Eifert H, Gunther B. Technegas: a study of particle structure, size and distribution. Eur J Nucl Med 1993;20:576–579. |
| 43. | Burch WM, Boyd MM, Crellin DE. Technegas: particle size and distribution. Eur J Nucl Med 1994;21:365–367. |
| 44. | Crawford A, Davison A, Amis T, Engel L. Intrapulmonary distribution of 99m technetium labelled ultrafine carbon aerosol (Technegas) in severe airflow obstruction. Eur Respir J 1990;3:686–692. |
| 45. | Amis T, Crawford A, Davison A, Engel L. Distribution of inhaled 99m technetium labelled ultrafine carbon particle aerosol (Technegas) in human lungs. Eur Respir J 1990;3:679–685. |
| 46. | Amis TC, Jones T. Krypton-81m as a flow tracer in the lung: theory and quantitation. Bull Eur Physiopathol Respir 1980;16:245–259. |
| 47. | Suga K. Technical and analytical advances in pulmonary ventilation SPECT with xenon-133 gas and Tc-99m-Technegas. Ann Nucl Med 2002;16:303–310. |
| 48. | Stavngaard T, Sogaard LV, Mortensen J, Hanson LG, Schmiedeskamp J, Berthelsen AK, Dirksen A. Hyperpolarised 3He MRI and 81mKr SPECT in chronic obstructive pulmonary disease. Eur J Nucl Med Mol Imaging 2004;32:448–457. |
