TY - JOUR
AU - Knaapen,, Paul
AB - Abstract Aims Despite high variability in coronary anatomy, quantitative positron emission tomography (PET) perfusion in coronary territories is traditionally calculated according to the American Heart Association (AHA) 17-segments model. This study aimed to assess the impact of individualized segmentation of myocardial segments on the diagnostic accuracy of hyperaemic myocardial blood flow (MBF) values for haemodynamically significant coronary artery disease (CAD). Methods and results Patients with suspected CAD (n = 204) underwent coronary computed tomography angiography (CCTA) and [15O]H2O PET followed by invasive coronary angiography with fractional flow reserve assessment of all major coronary arteries. Hyperaemic MBF per vascular territory was calculated using both standard segmentation according to the AHA model and individualized segmentation, in which CCTA was used to assign coronary arteries to PET perfusion territories. In 122 (59.8%) patients, one or more segments were redistributed after individualized segmentation. No differences in mean MBF values were seen between segmentation methods, except for a minor difference in hyperaemic MBF in the LCX territory (P = 0.001). These minor changes resulted in discordant PET-defined haemodynamically significant CAD between the two methods in only 5 (0.8%) vessels. The diagnostic value for detecting haemodynamically significant CAD did not differ between individualized and standard segmentation, with area under the curves of 0.79 and 0.78, respectively (P = 0.34). Conclusions Individualized segmentation using CCTA-derived coronary anatomy led to redistribution of standard myocardial segments in 60% of patients. However, this had little impact on [15O]H2O PET MBF values and diagnostic value for detecting haemodynamically significant CAD did not change. Therefore, clinical impact of individualized segmentation seems limited. positron emission tomography, segmentation, ischaemia, coronary artery disease, myocardial perfusion Introduction Positron emission tomography (PET) is a radionuclide imaging modality that allows for non-invasive quantification of myocardial blood flow (MBF).1 Among the myriad of available imaging techniques, quantitative PET currently yields the highest diagnostic accuracy for detecting obstructive coronary artery disease (CAD).2 In the assessment of MBF in specific vascular territories, the standard 17-segments American Heart Association (AHA) model is used to divide the left ventricular (LV) wall into the vascular territories of the left anterior descending artery (LAD), left circumflex artery (LCX), and right coronary artery (RCA).3 This standard model, however, is an oversimplification of the actual vascular architecture in a specific patient, since there is great variability in coronary anatomy. Variability in coronary anatomy is most frequently seen in difference in coronary dominance in which the posterior descending artery (PDA) is supplied by either the LCX or RCA and in second or third order vessels such as the anterolateral wall which may be supplied by diagonal or obtuse marginal branches. Individualized segmentation of the LV wall based on actual coronary anatomy would potentially lead to a more accurate assessment of vessel-specific MBF with a subsequent increase in diagnostic performance. Although several small studies have shown the feasibility of using coronary computed tomography angiography (CCTA) to obtain individualized segmentation in perfusion imaging, data on the diagnostic performance of this individualized method in quantitative PET perfusion imaging for detecting haemodynamically significant CAD are scarce.4–7 The purpose of this study was, therefore, to assess diagnostic accuracy of quantitative [15O]H2O PET perfusion measurements to detect haemodynamically significant CAD, as defined by fractional flow reserve (FFR), using both standard and individualized segmentations. Methods Study population The current report is a substudy of the PACIFIC trial and details regarding the study design are described previously.2 The study population consisted of 208 patients with stable new-onset chest pain and suspected CAD. All patients were aged 40 years and above and had intermediate pre-test probability for CAD as defined by Diamond and Forrester criteria. Major exclusion criteria were renal failure (i.e. estimated glomerular filtration rate <45 mL/min), a LV ejection fraction <50%, history of COPD or chronic asthma, a prior history of CAD, atrial fibrillation, and second or third degree AV block. In brief, patients underwent CCTA, single photon emission computed tomography, and [15O]H2O PET imaging, followed by invasive coronary angiography with FFR measurement in all major coronary vessels. For the present substudy all patients with PET imaging were included (n = 204). The study protocol was approved by the Medical Ethics Committee of the VU University Medical Center and written informed consent was obtained from all participants. Invasive coronary angiography and FFR Invasive coronary angiography and FFR measurements were performed as described previously.2 In brief, all major coronary arteries were routinely interrogated by FFR except for occluded or subtotal lesions with a diameter stenosis ≥90%. Maximal hyperaemia was induced by intracoronary (150 µg) or intravenous (140 µg/kg/min) administration of adenosine. An FFR of ≤0.80 was considered haemodynamically significant. In case FFR measurement was not performed, a stenosis of ≥90% was deemed significant, whereas a stenosis of ≤30% (obtained with Quantitative Coronary Analysis, QCA) was deemed non-significant.8 [15O]H2O PET acquisition PET imaging was performed on the same day as CCTA imaging. Patients were scanned using a hybrid PET-CT device (Philips Gemini TF 64, Philips Healthcare, Best, The Netherlands). The scanning protocol has been described in detail previously.9 In summary, a dynamic PET perfusion scan was performed during resting conditions using 370 MBq of [15O]H2O. A 6-min emission scan was started simultaneously with the administration of [15O]H2O. This dynamic scan sequence was followed immediately by a low-dose CT scan for attenuation correction. After a 10-min interval to allow for decay of radioactivity, an identical PET sequence was performed during hyperaemic conditions induced by intravenous adenosine infusion (140 µg/kg/min), initiated 2 min before the hyperaemic scan to ensure maximal vasodilation. CCTA acquisition Patient preparation and image acquisition were performed as described previously.2 In short, patients underwent CCTA on a 256-slice CT-scanner (Brilliance iCT, Philips Healthcare, Best, the Netherlands) with a collimation of 128 × 0.625 mm and a tube rotation time of 270 ms. Tube current was set between 200 and 360 mAs at 120 kV, adjusting primarily the mAs based on body habitus. Axial scanning was performed with prospective ECG-gating (Step & Shoot Cardiac, Philips Healthcare) at 75% of the R-R interval. A bolus of 100 mL iobitidol (Xenetix 350) was injected intravenously (5.7 mL/s) followed by a 50-mL saline flush. The scan was triggered using an automatic bolus tracking technique, with a region of interest placed in the descending thoracic aorta with a threshold of 150 Hounsfield units. In patients with a pre-scan heart rate of ≥65 bpm, metoprolol 50–150 mg was administered orally 1 h before the start of the CT protocol. If necessary, 5–25 mg metoprolol was given intravenously just before the scan to achieve a heart rate <65 bpm. All patients received 800 µg of sublingual nitroglycerine immediately before scanning. PET/CT analysis PET images were reconstructed using the 3D row action maximum likelihood algorithm and applying all appropriate corrections. Parametric MBF images were generated and quantitatively analysed using in-house developed software (Cardiac VUer).10 Myocardial blood flow was expressed in mL/min/g and both resting and hyperaemic MBF were calculated for the LV as a whole. Calculation of MBF values in the vascular territories of the LAD, LCX, and RCA was performed using two different segmentation methods (Figure 1). First, segmentation of the three vascular territories was derived from the 17-segment standard AHA model (standard segmentation). Second, CCTA was used to define the vascular territories of the three coronary arteries (individualized segmentation). For the individualized segmentation, short-axis projections of the LV were used. Four maximum intensity projection slices of equal thickness were used to assign each of the 17 myocardial segments to one of the three major coronary arteries (LAD, LCX, and RCA), hereby creating individualized polar maps.4 These individualized polar maps were then used to generate mean MBF values for the corresponding vascular territories. In both the standard and the individualized segmentation, PET-defined haemodynamically significant CAD was defined by a perfusion defect of at least two adjacent myocardial segments with hyperaemic MBF ≤2.30 mL/min/g.2,11 In vascular territories in which a perfusion defect was present, MBF and myocardial flow reserve (MFR) were defined by the mean of all segments with a perfusion defect. In vascular territories where no perfusion defect was present the MBF and MFR were calculated by the mean of all myocardial segments in that territory. Myocardial flow reserve was defined as the ratio between hyperaemic and resting MBF. Figure 1 View largeDownload slide Case example of myocardial segmentation analysis. CCTA images of a 61-year-old man presented with atypical chest pain. Standard segmentation was performed according to the AHA model (left panel). In this model, the left ventricle was divided into 17 segments and the LAD vascular territory was comprised of seven segments, whereas the LCX and RCA vascular territories were comprised of five segments. For the individualized method, four maximum intensity projection slices of equal thickness were generated to visualize the four short-axis views of the left ventricle (middle panel). These images were used to assign each of the 17 myocardial segments to one of the three major coronary arteries (LAD, LCX, and RCA), hereby creating an individualized segmentation of the left ventricle wall (right panel). In this case, there is a left-dominant coronary circulation in which the RCA was not subtending any left ventricular mass. Therefore segments 3, 4, 9, 10, and 15 were reassigned to the LCX. Furthermore, there was a large second diagonal branch which supplied the apical lateral segment of the left ventricle (segment 16), and therefore this segment was reassigned to the LAD. Figure 1 View largeDownload slide Case example of myocardial segmentation analysis. CCTA images of a 61-year-old man presented with atypical chest pain. Standard segmentation was performed according to the AHA model (left panel). In this model, the left ventricle was divided into 17 segments and the LAD vascular territory was comprised of seven segments, whereas the LCX and RCA vascular territories were comprised of five segments. For the individualized method, four maximum intensity projection slices of equal thickness were generated to visualize the four short-axis views of the left ventricle (middle panel). These images were used to assign each of the 17 myocardial segments to one of the three major coronary arteries (LAD, LCX, and RCA), hereby creating an individualized segmentation of the left ventricle wall (right panel). In this case, there is a left-dominant coronary circulation in which the RCA was not subtending any left ventricular mass. Therefore segments 3, 4, 9, 10, and 15 were reassigned to the LCX. Furthermore, there was a large second diagonal branch which supplied the apical lateral segment of the left ventricle (segment 16), and therefore this segment was reassigned to the LAD. In patients in whom CCTA showed a left-dominant coronary anatomy in which the RCA did not subtend any LV myocardium, the RCA vessel was excluded from the comparative analysis of the two segmentations methods for diagnostic value for FFR-defined haemodynamically significant CAD. The myocardial segments subtended by the PDA were attributed to the LCX territory in these cases. Statistical analyses All statistical analyses were performed using the SPSS software package (version 20.0.0, IBM SPSS Statistics, Armonk, NY, USA), except for receiver operating characteristics (ROC) curve analyses which were performed with MedCalc for Windows (version 12.7.8.0, MedCalc Software, Oostende, Belgium). Continuous variables were tested for normal distribution. Normal distributed continuous variables are presented as mean ± standard deviation. Non-normal distributed variables are presented as median with interquartile range. Categorical variables are presented as frequencies with percentages and compared with the χ2 test. For the separate vessels, the mean rest MBF, hyperaemic MBF, and MFR were compared between standard and individualized segmentation using a linear mixed model with a fixed effect for method of segmentation and random effects for patient. Sensitivity, specificity, accuracy, negative predictive value (NPV), and positive predictive value (PPV) for the presence of FFR-defined haemodynamically significant CAD averaged over all LAD, RCA, and LCX vessels were estimated and compared between the two segmentation methods using generalized estimating equations for dichotomous outcome with an unstructured correlation structure to account for correlation of observations made on vessels from the same patient. Vessel-type specific sensitivity, specificity, and accuracy were compared using the McNemar test and vessel-type specific NPV and PPV using generalized estimating equations an independent correlation structure.12 Furthermore, the respective areas under the ROC curves were compared using the method of DeLong for both per-vessel and per-patient analyses. A P-value <0.05 was considered statistically significant. Results Study population and LV segmentation The baseline characteristics of all 204 included patients are summarized in Table 1. Individualized segmentation of the LV wall resulted in redistribution of 287/3468 (8.3%) myocardial segments as compared with standard segmentation. In 122 (59.8%) patients, one or more segments were redistributed after individualized segmentation. A schematic overview of the distribution of specific myocardial segments supplied by each individual coronary artery (LAD, LCX, and RCA) according to the individualized segmentation is provided in Figure 2A. The relative occurrence of redistribution with the individualized method for each myocardial segment is depicted in Figure 2B. Redistribution of segments occurred in up to 29% in segments of the standard RCA and LCX territory. In contrast, redistribution did not occur in the standard LAD territory, except for one case of coronary anomaly (hypoplastic LAD) and in 5 (2.5%) cases in which the apex which was supplied by a different coronary. Redistribution occurred most often in the inferior (segments 4, 10, and 15) and lateral wall (segments 12 and 16). Figure 2 View largeDownload slide Individualized segmentation per-vessel and redistribution between standard and individualized segmentation. (A) Polar maps of the left ventricle visualizing the distribution of specific LV segments supplied by each individual coronary artery (LAD, LCX, and RCA, respectively). The colours indicate the proportion of cases in which the segment was supplied by that particular coronary artery (0–100%). (B) Polar map depicting the relative occurrence of redistribution of coronary segments with the individualized segmentation model. The colours indicate the frequency of redistribution of a particular segment (0–30%). Figure 2 View largeDownload slide Individualized segmentation per-vessel and redistribution between standard and individualized segmentation. (A) Polar maps of the left ventricle visualizing the distribution of specific LV segments supplied by each individual coronary artery (LAD, LCX, and RCA, respectively). The colours indicate the proportion of cases in which the segment was supplied by that particular coronary artery (0–100%). (B) Polar map depicting the relative occurrence of redistribution of coronary segments with the individualized segmentation model. The colours indicate the frequency of redistribution of a particular segment (0–30%). Table 1 Baseline characteristics Demographics N = 204 Age (years) 58 ± 9 Male, n (%) 129 (63) Body mass index 27 ± 4 Cardiovascular risk factors, n (%)  Diabetes mellitus type II 32 (16)  Hypertension 96 (47)  Hyperlipidaemia 81 (40)  Current tobacco use 40 (20)  History of tobacco use 99 (49)  Family history of CAD 104 (51) Type of chest pain, n (%)  Typical angina 71 (35)  Atypical angina 77 (38)  Non-specific chest discomfort 56 (28) Demographics N = 204 Age (years) 58 ± 9 Male, n (%) 129 (63) Body mass index 27 ± 4 Cardiovascular risk factors, n (%)  Diabetes mellitus type II 32 (16)  Hypertension 96 (47)  Hyperlipidaemia 81 (40)  Current tobacco use 40 (20)  History of tobacco use 99 (49)  Family history of CAD 104 (51) Type of chest pain, n (%)  Typical angina 71 (35)  Atypical angina 77 (38)  Non-specific chest discomfort 56 (28) CAD, coronary artery disease. View Large Table 1 Baseline characteristics Demographics N = 204 Age (years) 58 ± 9 Male, n (%) 129 (63) Body mass index 27 ± 4 Cardiovascular risk factors, n (%)  Diabetes mellitus type II 32 (16)  Hypertension 96 (47)  Hyperlipidaemia 81 (40)  Current tobacco use 40 (20)  History of tobacco use 99 (49)  Family history of CAD 104 (51) Type of chest pain, n (%)  Typical angina 71 (35)  Atypical angina 77 (38)  Non-specific chest discomfort 56 (28) Demographics N = 204 Age (years) 58 ± 9 Male, n (%) 129 (63) Body mass index 27 ± 4 Cardiovascular risk factors, n (%)  Diabetes mellitus type II 32 (16)  Hypertension 96 (47)  Hyperlipidaemia 81 (40)  Current tobacco use 40 (20)  History of tobacco use 99 (49)  Family history of CAD 104 (51) Type of chest pain, n (%)  Typical angina 71 (35)  Atypical angina 77 (38)  Non-specific chest discomfort 56 (28) CAD, coronary artery disease. View Large [15O]H2O PET An overview of mean rest MBF, mean hyperaemic MBF, and mean MFR values based on standard and individualized segmentation methods on a per-vessel level is summarized in Table 2. There were no significant differences in mean rest MBF and MFR values between the two segmentation methods. Hyperaemic MBF values were only significantly different in the LCX (P < 0.001), no differences in hyperaemic MBF were observed in the LAD (P = 0.75) and RCA (P = 0.82). In 14 patients, CCTA showed a left-dominant coronary anatomy in which the RCA did not subtend the LV myocardium. In these patients, calculation of hyperaemic MBF in the RCA territory based on individualized segmentation method was not possible, and these vessels were thus excluded from further analysis. Therefore, the total number of vessels included for further analysis of PET perfusion was 598 (97.7%). The relationship between hyperaemic MBF values calculated using standard and individualized segmentation methods is illustrated in Figure 3. There was a strong correlation between hyperaemic MBF calculated with the two segmentation methods (r = 0.98, P < 0.001; Figure 3A) and excellent agreement (ICC = 0.99, P < 0.001; Figure 3B). Bland–Altman plots showed an overall bias of 0.015 ± 0.15 mL/min/g (Figure 3B). Despite these minimal differences in MBF values, the presence of PET-defined haemodynamically significant CAD (hyperaemic MBF ≤ 2.3 mL/min/g in two adjacent myocardial segments) did not differ between the two segmentation methods. PET-defined haemodynamically significant CAD was present in 252 (42.1%) vessels in the standard method and 253 (42.3%) vessels in the individualized method (P = 0.95). PET-defined haemodynamically significant CAD results based on both segmentation methods were concordant in 593 (99.2%) vessels (Figure 3A). Discordance in PET-defined haemodynamically significant CAD between the two methods occurred in only 5 (0.8%) vessels: two vessels changed from haemodynamically significant to non-haemodynamically significant (left upper quadrant, Figure 3A) and three changed from non-haemodynamically significant to haemodynamically significant CAD with the use of the individualized segmentation (right lower quadrant, Figure 3A). Figure 3 View largeDownload slide Relationship between hyperaemic MBF derived from standard and individualized segmentation methods. (A) Scatterplot demonstrating a strong correlation between hyperaemic MBF values calculated using both standard and individualized segmentation methods (r = 0.99, P < 0.001). Concordance in MBF results between both methods (concordant negative result, right upper quadrant; concordant positive result, left lower quadrant) are depicted in blue. Discordant MBF measurements with a corresponding negative FFR measurement are depicted in green, whereas discordant measurements with a positive FFR measurement are depicted in red. (B) Bland–Altman plot demonstrating excellent agreement between hyperaemic MBF calculated with both segmentation methods (ICC = 0.99, P < 0.001). Figure 3 View largeDownload slide Relationship between hyperaemic MBF derived from standard and individualized segmentation methods. (A) Scatterplot demonstrating a strong correlation between hyperaemic MBF values calculated using both standard and individualized segmentation methods (r = 0.99, P < 0.001). Concordance in MBF results between both methods (concordant negative result, right upper quadrant; concordant positive result, left lower quadrant) are depicted in blue. Discordant MBF measurements with a corresponding negative FFR measurement are depicted in green, whereas discordant measurements with a positive FFR measurement are depicted in red. (B) Bland–Altman plot demonstrating excellent agreement between hyperaemic MBF calculated with both segmentation methods (ICC = 0.99, P < 0.001). Table 2 PET-derived myocardial blood flow parameters based on both standard and individualized segmentation methods Standard segmentation Individualized segmentation P-value Mean rest MBF  LAD territory 0.96 ± 0.27 0.97 ± 0.31 0.58  LCX territory 0.94 ± 0.25 0.94 ± 0.31 0.84  RCA territory 0.86 ± 0.24 0.87 ± 0.30* 0.73 Mean hyperaemic MBF  LAD territory 2.73 ± 1.25 2.73 ± 1.26 0.75  LCX territory 2.88 ± 1.19 2.83 ± 1.14 0.001  RCA territory 2.63 ± 1.17 2.64 ± 1.20* 0.82 Mean MFR  LAD territory 2.98 ± 1.24 2.98 ± 1.23 0.73  LCX territory 3.14 ± 1.22 3.12 ± 1.16 0.52  RCA territory 3.18 ± 1.41 3.18 ± 1.44* 0.34 Standard segmentation Individualized segmentation P-value Mean rest MBF  LAD territory 0.96 ± 0.27 0.97 ± 0.31 0.58  LCX territory 0.94 ± 0.25 0.94 ± 0.31 0.84  RCA territory 0.86 ± 0.24 0.87 ± 0.30* 0.73 Mean hyperaemic MBF  LAD territory 2.73 ± 1.25 2.73 ± 1.26 0.75  LCX territory 2.88 ± 1.19 2.83 ± 1.14 0.001  RCA territory 2.63 ± 1.17 2.64 ± 1.20* 0.82 Mean MFR  LAD territory 2.98 ± 1.24 2.98 ± 1.23 0.73  LCX territory 3.14 ± 1.22 3.12 ± 1.16 0.52  RCA territory 3.18 ± 1.41 3.18 ± 1.44* 0.34 The rest and stress MBF and MFR values represent the mean values of all 204 vessels in that particular vascular territory. The RCA territories in the individualized segmentation (*) contained only 190 vessels instead of 204 vessels, due to the presence of a left-dominant coronary system in 14 cases in which the RCA did not subtend any LV myocardium. In these cases, the myocardial segments subtended by the PDA were attributed to the LCX territory. MBF, myocardial blood flow; MFR, myocardial flow reserve; PET, positron emission tomography. View Large Table 2 PET-derived myocardial blood flow parameters based on both standard and individualized segmentation methods Standard segmentation Individualized segmentation P-value Mean rest MBF  LAD territory 0.96 ± 0.27 0.97 ± 0.31 0.58  LCX territory 0.94 ± 0.25 0.94 ± 0.31 0.84  RCA territory 0.86 ± 0.24 0.87 ± 0.30* 0.73 Mean hyperaemic MBF  LAD territory 2.73 ± 1.25 2.73 ± 1.26 0.75  LCX territory 2.88 ± 1.19 2.83 ± 1.14 0.001  RCA territory 2.63 ± 1.17 2.64 ± 1.20* 0.82 Mean MFR  LAD territory 2.98 ± 1.24 2.98 ± 1.23 0.73  LCX territory 3.14 ± 1.22 3.12 ± 1.16 0.52  RCA territory 3.18 ± 1.41 3.18 ± 1.44* 0.34 Standard segmentation Individualized segmentation P-value Mean rest MBF  LAD territory 0.96 ± 0.27 0.97 ± 0.31 0.58  LCX territory 0.94 ± 0.25 0.94 ± 0.31 0.84  RCA territory 0.86 ± 0.24 0.87 ± 0.30* 0.73 Mean hyperaemic MBF  LAD territory 2.73 ± 1.25 2.73 ± 1.26 0.75  LCX territory 2.88 ± 1.19 2.83 ± 1.14 0.001  RCA territory 2.63 ± 1.17 2.64 ± 1.20* 0.82 Mean MFR  LAD territory 2.98 ± 1.24 2.98 ± 1.23 0.73  LCX territory 3.14 ± 1.22 3.12 ± 1.16 0.52  RCA territory 3.18 ± 1.41 3.18 ± 1.44* 0.34 The rest and stress MBF and MFR values represent the mean values of all 204 vessels in that particular vascular territory. The RCA territories in the individualized segmentation (*) contained only 190 vessels instead of 204 vessels, due to the presence of a left-dominant coronary system in 14 cases in which the RCA did not subtend any LV myocardium. In these cases, the myocardial segments subtended by the PDA were attributed to the LCX territory. MBF, myocardial blood flow; MFR, myocardial flow reserve; PET, positron emission tomography. View Large Diagnostic accuracy of both segmentation methods for haemodynamically significant CAD FFR measurements were performed in 537 (89.8%) vessels. In 58 (9.6%) vessels, FFR was not performed due to a total or subtotal lesion. These vessels all showed >90% diameter stenosis and were deemed haemodynamically obstructed. Additionally, FFR measurement was not possible in 3 (0.5%) vessels because of severe tortuosity. None of these vessels showed a coronary lesion of ≥30% diameter stenosis and all were deemed non-obstructed. The total number of vessels with haemodynamically significant CAD was 160 (26.8%). Based on FFR results, four out of five vessels with discordance in PET-defined haemodynamically significant CAD between the two segmentation methods, were falsely reclassified by the individualized segmentation: two vessels changed from true positive to false negative (Figure 3A, left upper quadrant) and two vessels changed from true negative to false positive (Figure 3A, right lower quadrant). Only one vessel was adequately reclassified with the individualized method, from false negative to true positive (Figure 3A, right lower quadrant). The diagnostic performance of hyperaemic MBF with standard and individualized segmentation methods for detecting presence of haemodynamically significant CAD are summarized in Table 3 on a per-vessel and per-patient basis. No differences in sensitivity, specificity, NPV, PPV, and diagnostic accuracy were noted between the two groups. Furthermore, ROC curve analysis showed no significant difference in diagnostic value of hyperaemic MBF between standard and individualized segmentation methods, with AUCs of 0.79 in both methods in per-vessel analysis (P = 0.34) and AUCs of 0.82 and 0.83, respectively, in per-patient analysis (P = 0.33). The results of the impact of the two segmentation methods on the diagnostic performance of hyperaemic MBF in the subgroups of the individual coronary arteries (LAD, LCX, and RCA) are shown in the Supplementary data online, Table S1. There were no significant differences in diagnostic value between the two segmentation methods in any of the individual coronary arteries. Table 3 Diagnostic accuracy of hyperaemic MBF for detecting haemodynamically significant CAD using both standard and individualized segmentation methods Standard segmentation Individualized segmentation P-value Per-vessel  Sensitivity (%) 84 (76–90) 83 (75–89) 0.38  Specificity (%) 69 (62–75) 69 (62–75) 0.20  NPV (%) 94 (90–96) 93 (89–95) 0.60  PPV (%) 54 (46–61) 54 (47–61) 0.73  Accuracy (%) 77 (72–81) 76 (71–80) 0.34  AUC 0.79 (0.76–0.82) 0.79 (0.75–0.82) 0.34 Per-patient  Sensitivity (%) 87 (78–93) 88 (80–94) 1.00  Specificity (%) 78 (69–85) 78 (69–85) 1.00  NPV (%) 88 (80–94) 89 (81–94) 1.00  PPV (%) 76 (67–84) 76 (67–84) 0.32  Accuracy (%) 82 (76–87) 82 (76–87) 0.32  AUC 0.82 (0.76–0.87) 0.83 (0.77–0.88) 0.32 Standard segmentation Individualized segmentation P-value Per-vessel  Sensitivity (%) 84 (76–90) 83 (75–89) 0.38  Specificity (%) 69 (62–75) 69 (62–75) 0.20  NPV (%) 94 (90–96) 93 (89–95) 0.60  PPV (%) 54 (46–61) 54 (47–61) 0.73  Accuracy (%) 77 (72–81) 76 (71–80) 0.34  AUC 0.79 (0.76–0.82) 0.79 (0.75–0.82) 0.34 Per-patient  Sensitivity (%) 87 (78–93) 88 (80–94) 1.00  Specificity (%) 78 (69–85) 78 (69–85) 1.00  NPV (%) 88 (80–94) 89 (81–94) 1.00  PPV (%) 76 (67–84) 76 (67–84) 0.32  Accuracy (%) 82 (76–87) 82 (76–87) 0.32  AUC 0.82 (0.76–0.87) 0.83 (0.77–0.88) 0.32 All values except AUC are % (95% confidence interval). AUC, area under the curve; NPV, negative predictive value; PPV, positive predictive value. View Large Table 3 Diagnostic accuracy of hyperaemic MBF for detecting haemodynamically significant CAD using both standard and individualized segmentation methods Standard segmentation Individualized segmentation P-value Per-vessel  Sensitivity (%) 84 (76–90) 83 (75–89) 0.38  Specificity (%) 69 (62–75) 69 (62–75) 0.20  NPV (%) 94 (90–96) 93 (89–95) 0.60  PPV (%) 54 (46–61) 54 (47–61) 0.73  Accuracy (%) 77 (72–81) 76 (71–80) 0.34  AUC 0.79 (0.76–0.82) 0.79 (0.75–0.82) 0.34 Per-patient  Sensitivity (%) 87 (78–93) 88 (80–94) 1.00  Specificity (%) 78 (69–85) 78 (69–85) 1.00  NPV (%) 88 (80–94) 89 (81–94) 1.00  PPV (%) 76 (67–84) 76 (67–84) 0.32  Accuracy (%) 82 (76–87) 82 (76–87) 0.32  AUC 0.82 (0.76–0.87) 0.83 (0.77–0.88) 0.32 Standard segmentation Individualized segmentation P-value Per-vessel  Sensitivity (%) 84 (76–90) 83 (75–89) 0.38  Specificity (%) 69 (62–75) 69 (62–75) 0.20  NPV (%) 94 (90–96) 93 (89–95) 0.60  PPV (%) 54 (46–61) 54 (47–61) 0.73  Accuracy (%) 77 (72–81) 76 (71–80) 0.34  AUC 0.79 (0.76–0.82) 0.79 (0.75–0.82) 0.34 Per-patient  Sensitivity (%) 87 (78–93) 88 (80–94) 1.00  Specificity (%) 78 (69–85) 78 (69–85) 1.00  NPV (%) 88 (80–94) 89 (81–94) 1.00  PPV (%) 76 (67–84) 76 (67–84) 0.32  Accuracy (%) 82 (76–87) 82 (76–87) 0.32  AUC 0.82 (0.76–0.87) 0.83 (0.77–0.88) 0.32 All values except AUC are % (95% confidence interval). AUC, area under the curve; NPV, negative predictive value; PPV, positive predictive value. View Large Discussion Our study evaluated the effect of individualized LV wall segmentation on the diagnostic value of quantitative [15O]H2O PET. Results show that although redistribution of myocardial segments from standard to individualized segmentation occurred in 60% of patients, impact on MBF values was small. Discordance in PET-defined haemodynamically significant CAD between standard and individualized segmentation occurred in only 0.8% of vessels. As a result, the diagnostic value for the presence of haemodynamically significant CAD did not change with the use of individualized segmentation. Therefore, there is no need to use coronary anatomy from CCTA to define vascular territories in clinical PET perfusion imaging. Correspondence of myocardial segments with coronary anatomy Since no information on coronary anatomy and lesion location is obtained with PET, the standard AHA 17-segments model is commonly used to assign individual LV perfusion segments to specific coronary arteries.3 This model is based on SPECT data obtained during transient coronary occlusion.13 Although widely adopted, the AHA model ignores the tremendous variability in coronary anatomy. An individualized segmentation method in which information on coronary anatomy is used to assign coronary arteries to perfusion territories has been proposed to overcome this shortcoming.4–7 In the current report, CCTA-based individualized segmentation resulted in redistribution of LV segments in 60% of patients and in 8% of segments. Segments in the LAD territory were almost completely specific for the LAD, whereas redistribution was frequent in the LCX and RCA territories. Redistribution occurred most often in the inferior (segments 4, 10, and 15) and lateral wall (segments 12 and 16). In general, these results are in line with prior findings.4–7,14–16 Most studies, like the present, have used anatomical information from CCTA to match myocardial segments with their feeding coronary artery.4–7 Pereztol-Valdés et al.16 and Ortiz-Pérez et al.,15 however, matched angiographic occlusion sites with, respectively, SPECT perfusion defects and cardiac magnetic resonance hyperenhancement. All prior studies, except Javadi et al., corroborate the present findings that segments of the standard LAD territory are almost exclusively attributed to the LAD. Furthermore, in line with the present findings, variability of coronary anatomy was largest in lateral myocardial segments (12 and 16). To some extent, variability in the inferior segments was also reported in all aforementioned studies. Interestingly, studies that used invasive angiography to obtain anatomic information noted high variability in the septal segments.15,16 This was not seen in our study, nor in most other CCTA studies.4,6 A possible explanation for this discrepancy is the fact that visualization of small intramural arteries to septal segments with CCTA is challenging. This may have led to underestimation of the arterial supply of the inferoseptal segments by the LAD. Another explanation could be that in invasive angiography the push of dye into a coronary artery may lead to visualization of vessels in watershed regions. Differences in the force of dye injection could therefore lead to under or overestimation of the vascular supply to the inferoseptal segments. Diagnostic value of quantitative PET with individualized segmentation Data on the impact of CCTA-defined individualized segmentation on the diagnostic value of PET perfusion imaging are scarce. Liga et al.7 reported on the diagnostic value of hybrid imaging with fusion of CCTA and perfusion images in 252 patients. In this predominantly SPECT/CT study, individualized segmentation resulted in a change in SPECT/PET-defined haemodynamically significant CAD in 18% of patients and an increased diagnostic accuracy of hybrid vs. stand-alone imaging. The validity of these results for PET imaging however are unclear, as only 29% of patients underwent PET/CT and no data on the PET/CT subgroup were presented. Furthermore, the increased diagnostic value may not solely be the result of differences in segmentation, as the addition of CCTA-derived stenosis grade may have also augmented diagnostic performance. Therefore, no reliable conclusion on the sole effect of individualized segmentation can be drawn from this study. To date, only one study has investigated the effect of individualized segmentation on the diagnostic value of quantitative PET perfusion. Thomassen et al.4 reported no difference in hyperaemic MBF values between standard and individualized segmentation, in a small population of 44 patients undergoing [15O]H2O PET and CCTA. As a consequence, individualized segmentation of myocardial segments had little effect on PET-defined haemodynamically significant CAD results with discordance in PET-defined haemodynamically significant CAD between segmentation methods in only one patient. Using QCA stenosis ≥50% as reference standard, no difference in diagnostic performance between the segmentation methods was noted. In contrast with Thomassen et al., redistribution of segments by the individualized method led to minimal changes in hyperaemic MBF values in the current study. These changes, however, were very small and there was excellent agreement of MBF values between the two methods (ICC = 0.99). In concordance with Thomassen et al., discordance in PET-defined haemodynamically significant CAD between the segmentation methods was very infrequent and occurred in only 5 (0.8%) vessels. Only one of these discordant vessels was correctly reclassified according to FFR. As a result, the diagnostic value of [15O]H2O PET for detecting obstructive CAD did not change with the use of individualized segmentation. Although similar to Thomassen et al., our results expand on their findings in three ways. First, Thomassen et al. reported almost no redistribution of inferior myocardial segments. Since this is likely an effect of the relatively small patient population,17 the influence of individualized segmentation on the diagnostic value reported in their study may not adequately reflect the effect in larger patient populations. Secondly, the reference standard for haemodynamically significant CAD of QCA-defined stenosis is considered suboptimal, since several studies have shown that angiographic stenosis severity does not adequately predict the functional consequences of a stenosis.8 Therefore, FFR was used as reference for haemodynamically significant CAD in our study. Thirdly, Thomassen et al. used mean hyperaemic MBF value in the entire perfusion territory to define haemodynamically significant CAD. Since perfusion defects seldom cover the whole perfusion territory, MBF values will be averaged over both ischaemic and non-ischaemic segments. This may lead to underestimation of haemodynamically significant CAD. Therefore, we used the previously validated and commonly used definition of two adjacent myocardial segments with a hyperaemic MBF ≤2.3 for PET-defined haemodynamically significant CAD.2,11 Using these definitions for PET and FFR, diagnostic accuracy of [15O]H2O PET was not altered by the individualized segmentation of myocardial segments. Therefore, there is no need to perform the time-consuming procedure of matching CCTA-defined coronary anatomy with PET perfusion territories in clinical PET perfusion imaging. Nevertheless, it is important to note that standard myocardial segments are an oversimplification of perfusion territories since single myocardial segments can be supplied by multiple coronary arteries. MBF values per segment may therefore be averaged over both ischaemic and non-ischaemic myocardium. Recent advances in CCTA analysis might enable more accurate delineation of perfusion defects. The Voronoi-algorithm is a mathematical algorithm which enables estimation of subtending myocardial mass from CCTA data.18–20 Prior studies have reported excellent correlation between the Voronoi-algorithm segmentation and both SPECT-based myocardial mass at risk in humans and histological myocardial mass in animal models.18–20 Furthermore, recent developments have enabled CT-perfusion-based MBF quantification in the vascular territory of a specific coronary stenosis using the Voronoi-algorithm.21 However, fusion of PET perfusion data with Voronoi-algorithm segmentation is not yet available. These technical advances in PET perfusion imaging have the potential to improve diagnostic performance and are eagerly awaited. Limitations This study has several limitations. First, as already mentioned, it is likely that arterial supply of septal segments by the LAD has been underestimated by the present CCTA-based segmentation method. This may have affected the diagnostic value of the individualized method. Second, FFR was not available in 61 (10.2%) vessels. In these vessels, assumptions on haemodynamically significant CAD were made. Although prior data have shown FFR to be positive in almost all vessels with a subtotal stenosis and negative in all vessels without ≥30% stenosis, these assumptions may have affected our results.8 Third, according to the AHA 17-segments model, the apex (segment 17) is defined as the area of myocardium beyond the end of the LV cavity.3 Instead, we used the methodology of Thomassen et al.4 in which the LV myocardium is divided into four slices of equal thickness. Although the differences between these segmentation methodologies are small, minor influence on our results may have occurred. Fourth, [15O]H2O was used as tracer in the current study. Since [15O]H2O is metabolically inert and freely diffusible across myocyte membrane and subsequently uptake in/and clearance from the myocardium is linear to perfusion, it is considered the optimal tracer for perfusion imaging.1 Although one should be cautious to directly extrapolated our results to PET perfusion imaging with other tracers, such as 13NH3 and 82Rb, it seems plausible that our conclusion may also hold true for imaging with other tracers. Fifth, several studies have reported on the clinical impact of hybrid imaging with fusion of CCTA and perfusion images which enables co-localization of myocardial perfusion abnormalities and subtending coronary arteries.22,23 This might result in more precise adjudication of perfusion territories than with separate interpretation of CT and PET/SPECT. In our study, CCTA and PET analysis were performed in a blinded fashion to evaluate the sole effect of myocardial segmentation on the diagnostic value of quantitative PET perfusion. Individualized segmentation with image fusion might have yielded different results. Sixth, the impact of individualized segmentation may be influenced by knowledge of the complexity of disease. Since analysis of PET data was performed blinded for CCTA and angiographic data, this will not have affected our results. Last, the added value of the use of CCTA for segmentation of myocardial segments might theoretically be larger in inexperienced readers. Since PET imaging analysis in the current study was performed by an experienced core laboratory, our results may not be extrapolated to the diagnostic value of PET when evaluated by inexperienced readers. Conclusion Individualized segmentation of myocardial segments using CCTA-derived coronary anatomy led to redistribution of standard myocardial segments in 60% of patients. However, the impact on [15O]H2O PET-derived MBF values was small, with discordance in PET-defined haemodynamically significant CAD between standard and individualized segmentation in only 0.8% of vessels. As a result, the diagnostic value of [15O]H2O PET in detecting presence of haemodynamically significant CAD did not change with the use of individualized segmentation. Therefore, there is no need to use CCTA to allocate vascular territories to their corresponding coronary arteries in clinical PET perfusion imaging. Conflict of interest: J.K. reports receiving support from the Academy of Finland Centre of Excellence in Molecular Imaging in Cardiovascular and Metabolic Research, Helsinki, Finland, and receiving grant support from Gilead, Inc. and serving as a consultant to Lantheus, Inc. A.A.L. reports receiving research grants from AVID, Philips Healthcare, F. Hoffmann-La Roche Ltd., and the European Commission. References 1 Driessen RS , Raijmakers PG , Stuijfzand WJ , Knaapen P. Myocardial perfusion imaging with PET . Int J Cardiovasc Imaging 2017 ; 33 : 1021 – 31 . Google Scholar Crossref Search ADS PubMed 2 Danad I , Raijmakers PG , Driessen RS , Leipsic J , Raju R , Naoum C et al. Comparison of coronary CT angiography, SPECT, PET, and hybrid imaging for diagnosis of ischemic heart disease determined by fractional flow reserve . JAMA Cardiol 2017 ; 2 : 1100 – 7 . Google Scholar Crossref Search ADS PubMed 3 Cerqueira MD , Weissman NJ , Dilsizian V , Jacobs AK , Kaul S , Laskey WK et al. Standardized myocardial segmentation and nomenclature for tomographic imaging of the heart. A statement for healthcare professionals from the Cardiac Imaging Committee of the Council on Clinical Cardiology of the American Heart Association . Circulation 2002 ; 105 : 539 – 42 . Google Scholar Crossref Search ADS PubMed 4 Thomassen A , Petersen H , Johansen A , Braad PE , Diederichsen AC , Mickley H et al. Quantitative myocardial perfusion by O-15-water PET: individualized vs. standardized vascular territories . Eur Heart J Cardiovasc Imaging 2015 ; 16 : 970 – 6 . Google Scholar PubMed 5 Javadi MS , Lautamaki R , Merrill J , Voicu C , Epley W , McBride G et al. Definition of vascular territories on myocardial perfusion images by integration with true coronary anatomy: a hybrid PET/CT analysis . J Nucl Med 2010 ; 51 : 198 – 203 . Google Scholar Crossref Search ADS PubMed 6 Setser RM , O’Donnell TP , Smedira NG , Sabik JF , Halliburton SS , Stillman AE et al. Coregistered MR imaging myocardial viability maps and multi-detector row CT coronary angiography displays for surgical revascularization planning: initial experience . Radiology 2005 ; 237 : 465 – 73 . Google Scholar Crossref Search ADS PubMed 7 Liga R , Vontobel J , Rovai D , Marinelli M , Caselli C , Pietila M et al. Multicentre multi-device hybrid imaging study of coronary artery disease: results from the EValuation of INtegrated Cardiac Imaging for the detection and characterization of ischaemic heart disease (EVINCI) hybrid imaging population . Eur Heart J Cardiovasc Imaging 2016 ; 17 : 951 – 60 . Google Scholar Crossref Search ADS PubMed 8 Tonino PA , Fearon WF , De Bruyne B , Oldroyd KG , Leesar MA , Ver Lee PN et al. Angiographic versus functional severity of coronary artery stenoses in the FAME study fractional flow reserve versus angiography in multivessel evaluation . J Am Coll Cardiol 2010 ; 55 : 2816 – 21 . Google Scholar Crossref Search ADS PubMed 9 Danad I , Raijmakers PG , Harms HJ , Heymans MW , van Royen N , Lubberink M et al. Impact of anatomical and functional severity of coronary atherosclerotic plaques on the transmural perfusion gradient: a [15O]H2O PET study . Eur Heart J 2014 ; 35 : 2094 – 105 . Google Scholar Crossref Search ADS PubMed 10 Harms HJ , Knaapen P , de Haan S , Halbmeijer R , Lammertsma AA , Lubberink M. Automatic generation of absolute myocardial blood flow images using [15O]H2O and a clinical PET/CT scanner . Eur J Nucl Med Mol Imaging 2011 ; 38 : 930 – 9 . Google Scholar Crossref Search ADS PubMed 11 Danad I , Uusitalo V , Kero T , Saraste A , Raijmakers PG , Lammertsma AA et al. Quantitative assessment of myocardial perfusion in the detection of significant coronary artery disease: cutoff values and diagnostic accuracy of quantitative [(15)O]H2O PET imaging . J Am Coll Cardiol 2014 ; 64 : 1464 – 75 . Google Scholar Crossref Search ADS PubMed 12 Leisenring W , Alono T , Pepe MS. Comparisons of predictive values of binary medical diagnostic tests for paired designs . Biometrics 2000 ; 56 : 345 – 51 . Google Scholar Crossref Search ADS PubMed 13 Gallik DM , Obermueller SD , Swarna US , Guidry GW , Mahmarian JJ , Verani MS. Simultaneous assessment of myocardial perfusion and left ventricular function during transient coronary occlusion . J Am Coll Cardiol 1995 ; 25 : 1529 – 38 . Google Scholar Crossref Search ADS PubMed 14 Aepfelbacher FC , Johnson RB , Schwartz JG , Chen L , Parker RA , Parker JA et al. Validation of a model of left ventricular segmentation for interpretation of SPET myocardial perfusion images . Eur J Nucl Med 2001 ; 28 : 1624 – 9 . Google Scholar Crossref Search ADS PubMed 15 Ortiz-Perez JT , Rodriguez J , Meyers SN , Lee DC , Davidson C , Wu E. Correspondence between the 17-segment model and coronary arterial anatomy using contrast-enhanced cardiac magnetic resonance imaging . JACC Cardiovasc Imaging 2008 ; 1 : 282 – 93 . Google Scholar Crossref Search ADS PubMed 16 Pereztol-Valdes O , Candell-Riera J , Santana-Boado C , Angel J , Aguade-Bruix S , Castell-Conesa J et al. Correspondence between left ventricular 17 myocardial segments and coronary arteries . Eur Heart J 2005 ; 26 : 2637 – 43 . Google Scholar Crossref Search ADS PubMed 17 Knuuti J , Saraste A. Perfusion imaging and coronary anatomy . Eur Heart J Cardiovasc Imaging 2015 ; 16 : 966 – 7 . Google Scholar PubMed 18 Kurata A , Kono A , Sakamoto T , Kido T , Mochizuki T , Higashino H et al. Quantification of the myocardial area at risk using coronary CT angiography and Voronoi algorithm-based myocardial segmentation . Eur Radiol 2015 ; 25 : 49 – 57 . Google Scholar Crossref Search ADS PubMed 19 Ide S , Sumitsuji S , Yamaguchi O , Sakata Y. Cardiac computed tomography-derived myocardial mass at risk using the Voronoi-based segmentation algorithm: a histological validation study . J Cardiovasc Comput Tomogr 2017 ; 11 : 179 – 82 . Google Scholar Crossref Search ADS PubMed 20 Chung MS , Yang DH , Kim YH , Kang SJ , Jung J , Kim N et al. Myocardial segmentation based on coronary anatomy using coronary computed tomography angiography: development and validation in a pig model . Eur Radiol 2017 ; 27 : 4044 – 53 . Google Scholar Crossref Search ADS PubMed 21 Tomizawa N , Fujino Y , Kamitani M , Chou S , Yamamoto K , Inoh S et al. Longer diabetes duration reduces myocardial blood flow in remote myocardium assessed by dynamic myocardial CT perfusion . J Diabetes Complications 2018 ; 32 : 609 – 15 . Google Scholar Crossref Search ADS PubMed 22 Gaemperli O , Schepis T , Valenta I , Husmann L , Scheffel H , Duerst V et al. Cardiac image fusion from stand-alone SPECT and CT: clinical experience . J Nucl Med 2007 ; 48 : 696 – 703 . Google Scholar Crossref Search ADS PubMed 23 Santana CA , Garcia EV , Faber TL , Sirineni GK , Esteves FP , Sanyal R et al. Diagnostic performance of fusion of myocardial perfusion imaging (MPI) and computed tomography coronary angiography . J Nucl Cardiol 2009 ; 16 : 201 – 11 . Google Scholar Crossref Search ADS PubMed Published on behalf of the European Society of Cardiology. All rights reserved. © The Author(s) 2018. For permissions, please email: journals.permissions@oup.com. This article is published and distributed under the terms of the Oxford University Press, Standard Journals Publication Model (https://academic.oup.com/journals/pages/open_access/funder_policies/chorus/standard_publication_model)
TI - Impact of individualized segmentation on diagnostic performance of quantitative positron emission tomography for haemodynamically significant coronary artery disease
JO - European Heart Journal - Cardiovascular Imaging
DO - 10.1093/ehjci/jey201
DA - 2019-05-01
UR - https://www.deepdyve.com/lp/oxford-university-press/impact-of-individualized-segmentation-on-diagnostic-performance-of-CIpFGYJkV6
SP - 525
VL - 20
IS - 5
DP - DeepDyve
ER -