Incremental diagnostic value of perivascular fat attenuation index for identifying hemodynamically significant ischemia with severe calcification

Purpose: To explore the incremental value of perivascular fat attenuation index (FAI) to identify hemodynamically significant ischemia in severe calcified vessels. Methods: Patients who underwent coronary computed tomographic angiography (CCTA) examination at Chinese PLA General Hospital from 2017 to 2020 and subsequently underwent fractional flow reserve (FFR) examination within 1 month were consecutively included. Several CCTA-derived indices were measured, including the coronary artery calcification score (CACS), lesion length, ≥CAD-RADS 4 proportion, perivascular FAI and CT-FFR. The included vessels were divided into a nonsevere calcification group and a severe calcification group according to the quartile of CACS. FFR ≤ 0.80 represents the presence of hemodynamically significant ischemia. Results: A total of 124 patients with 152 vessels were included (age: 61.1 ± 9.2 years; male 64.5%). Significant differences in lesion length (28.4 ± 14.2 vs. 23.1 ± 12.3 mm, P = 0.021), perivascular FAI (-73.0 ± 7.5 vs. -79.0 ± 7.4 HU, P < 0.001) and CT-FFR (0.78 ± 0.06 vs. 0.86 ± 0.04, P < 0.001) were noted between the FFR ≤ 0.80 group (47 vessels) and the FFR > 0.80 group (105 vessels). Furthermore, the perivascular FAI in the FFR ≤ 0.80 group was significantly greater than that in the FFR > 0.80 group (nonsevere calcification: -73.2 ± 7.5 vs. -78.2 ± 7.4 HU, P = 0.002; severe calcification: -72.8 ± 7.7 vs. -82.7 ± 6.3 HU, P < 0.001). In discriminating hemodynamically significant ischemia, the specificity and accuracy of CT-FFR were significantly affected by severe calcification, which demonstrated a significantly declining trend (P = 0.033 and P = 0.010, respectively). The diagnostic performance of CT-FFR in the severe calcification group was lower than that in the nonsevere calcified group. However, perivascular FAI showed good discriminative performance in the severe calcification group. In combination with perivascular FAI, the predictive value of CT-FFR in identifying hemodynamically significant ischemia with severe calcification increased from an AUC of 0.740 to 0.919. Conclusion: For coronary artery with severe calcification, the diagnostic performance of CT-FFR in discriminating flow-limiting lesions could be greatly impaired. Perivascular FAI represents a potential reliable imaging marker to provide incremental diagnostic value over CT-FFR for identifying hemodynamically significant ischemia with severe calcification.


Introduction
The accuracy of noninvasive coronary computed tomographic angiography (CCTA) in the quantitative evaluation of coronary stenosis may be greatly impaired by severe coronary calcified lesions. When this examination is used in vessels with severe calcification, the accuracy and specificity of CCTA diagnosis will be decreased because of overestimation of the severity of stenosis [1,2]. Fractional flow reserve based on CCTA (CT-FFR) has emerged as an advanced noninvasive method for detecting coronary hemodynamic abnormalities developed in recent years [3,4]. CT-FFR can complement the usage of conventional CCTA, provide promising information for downstream clinical decisionmaking, and fulfill anatomical and functional assessments in a "one-stop" manner without additional procedures. A series of clinical registry trials have confirmed that combination with CT-FFR is superior to CCTA alone for the diagnosis of obstructive coronary artery disease (CAD) [5][6][7][8]. However, CT-FFR technology could be also impacted by the presence of artifacts, including motion, high image noise, and excessive calcium blooming [8,9]. Most importantly, manually labeling the physiological artery model and recognizing the lumen boundary remain necessary steps of CT-FFR measurement.
In recent years, it was reported that the perivascular fat attenuation index (FAI) may serve as a new imaging marker of coronary artery inflammation, which can be evaluated on CCTA reconstructive images [10]. The perivascular FAI around culprit lesions in ACS was greater than that in nonculprit lesions [10]. Combined with the total plaque volume and diameter stenosis rate, the perivascular FAI showed good diagnostic accuracy in predicting the hemodynamic significance of coronary stenosis [11]. Notably, elevated perivascular FAI was associated with increased adverse cardiovascular outcomes, which made it possible to use the novel marker to assess cardiovascular risk [12]. Considering that perivascular fat was outside the coronary artery and minimally influenced by calcification of the intimal layer, the application of perivascular FAI may have certain value for patients with severe calcified coronary arteries. The aim of this study was to explore the incremental diagnostic value of perivascular FAI over CT-FFR to identify hemodynamically significant CAD in severe calcified vessels.

Study population
Patient data were obtained from a retrospective cohort of symptomatic patients suspected of having CAD at Chinese PLA General Hospital from January 2017 to December 2020. Patients who underwent CCTA before ICA and FFR examination within a period of 1 month were consecutively included. ICA was driven by positive lesion of CCTA. Image of poor quality, which was insufficient to analyze CT-FFR or perivascular FAI, was not eligible for this study. Patients with a history of percutaneous coronary intervention or coronary artery bypass graft surgery were also excluded. Finally, 124 patients with 152 vessels who underwent FFR examination were eligible for inclusion. This study was approved by the Ethics committee of Chinese PLA General Hospital without the need for informed consent from all patients given the retrospective nature of the study.

Clinical cardiac risk factors
Basic clinical characteristics, including age, sex, body mass index (BMI), cardiac risk factors, including hypertension, diabetes mellitus, hyperlipidemia, and current smoking, were collected systematically. Hypertension was defined as systolic blood pressure ≥ 140 mmHg, diastolic blood pressure ≥ 90 mmHg or treatment with antihypertensive drugs. Diabetes mellitus was defined as fasting plasma glucose ≥ 7.0 mmol/L, 2-h plasma glucose ≥ 11.1 mmol/L during oral glucose tolerance test, HbA1C ≥ 6.5% (48 mmol/ mol), or a previous diagnosis of diabetes (medical diagnosis or the use of insulin/oral hypoglycemic agents). Hyperlipidemia was defined as serum total cholesterol ≥ 230 mg/ dL, serum triglycerides ≥ 200 mg/dL or a documented history of dyslipidemia (including the use of lipid-lowering medication).

CCTA protocol
All included patients underwent CCTA scans using a second-generation dual-source CT scanner (Somatom Definition Flash, Siemens Medical Solutions, Forchheim, Germany). Electrocardiography was continuously monitored throughout the entire scan. Patients were intravenously administered a 50 ~ 100 mg esmolol hydrochloride injection when the heart rate (HR) was > 70 beats/min and were instructed to hold their breath before the scan to avoid breathing artifacts. Nitroglycerin was sublingually administered to patients to ensure the dilation of vessels unless contraindicated. Before CCTA scanning, noncontrast CT was performed to measure the coronary artery calcification score (CACS) and to locate the scan range of the heart. Nonionic contrast medium (Ultravist®, 370 mgI/ml, Schering AG, Guangzhou, China) was injected intravenously into the antecubital vein at 5.5 ml/s. The region of interest was placed within the ascending aorta, and scanning was started when the attenuation was 100 HU greater than the baseline.

CT image analysis
All CCTA images were transferred to a workstation (Syngo. via VB10B, Siemens Healthcare, Germany) for analysis by two experienced investigators with > 10 years' experience who were blinded to the patients' clinical and scanning data. Coronary vessels with a diameter ≥ 2.0 mm were analyzed [13]. The degree of lumen stenosis ≥ 70% was represented as CAD-RADS system on pre vessel level, which could be recorded as CAD-RADS 4 [14]. Calcification was defined as 3 adjacent pixels > 130 HU. All calcified lesions in each coronary artery were labeled, and the CACS was calculated using the standard Agatston method [15]. Based on the quartile of CACS, the vessels were divided into a nonsevere calcification group (1st ~ 3rd quartiles, Q1 ~ Q3) and a severe calcification group (4th quartile, Q4).

Perivascular FAI measurement
Perivascular FAI was measured for all vessels interrogated with FFR examination on a dedicated software workstation (Anythink CT, Coronary Artery Analysis, version 1.01, CREALIFE, China). Perivascular fat tissue and FAI measurement were assessed using the methods previously described by Antonopoulos et al. [10]. Specifically, perivascular fat tissue was defined as the fat tissue within a radial distance from the outer vessel wall equal to the diameter of the vessel along the plaques (Fig. 1). For tandem lesions (distance > 1 cm) observed in one major coronary vessel, the perivascular FAI of the most severe lesion was measured. For the tandem lesions (distance ≤ 1 cm) or single lesion observed in one major coronary vessel, the segments were analyzed across the lesion between the proximal normal segment and the distal normal segment. The left main artery was not included for analysis due to a lack of FFR data. Perivascular FAI was quantified as an attenuation histogram of fat tissue within the range -190~-30 HU. The areas encasing small branches or coronary veins were manually excluded to avoid the effect of nonadipose tissue [11]. Disagreements between two investigators were resolved by consensus reading.

CT-FFR measurement
CT-FFR based on machine learning (ML) of all vessels interrogated with FFR examination was measured using artificial intelligence software (DeepFFR V1.0.0, Beijing CuraCloud Technology Co., Ltd., Beijing, China). The workstation included dedicated software utilizing the original CCTA imaging to simulate FFR values in an artificial intelligence model, which has been introduced in previous articles [16][17][18]. CT-FFR measurement of all patients was performed at the time of study analysis. The calculation process can be summarized as follows: (1) A deep learning algorithm was used to establish a characteristic sample database of coronary hemodynamic characteristic parameters.
(2) A 3D coronary artery model was extracted, and coronary centerlines were generated. The modified 3D U-Net-like model was employed to generate a major coronary artery tree followed by a graph cut to refine the boundary of the arteries. The centerlines were extracted using a minimal path extraction filter. (3) Then, the novel path-based deep learning model was used to predict the simulated FFR values. The whole model can process variable-length input, and each point of the input sequence was transferred separately corresponding to a multilayer perceptron network. Then, the output of the multilayer perceptron network was transferred (PPV) and negative predictive value (NPV) were calculated. Comparisons of diagnostic performances among calcification groups were performed using the chi-square test, Fisher's exact test, or McNemar's test. The AUC was compared using the DeLong method [19]. In statistical analysis, CT-FFR and perivascular FAI were mainly used as continuous variables, for ROC curve plotting and the calculation of AUC, as well as combined ROC curve plotting. When calculating diagnostic indicators of accuracy, sensitivity, specificity, PPV, and NPV, because 0.80 is wide accepted cutoff value of CT-FFR, this value was applied to convert CT-FFR into two categories; However, perivascular FAI has no recognized cutoff value, so the cutoff value calculated by the Youden index from ROC analysis when calculating sensitivity, specificity. All statistical analyses were performed using SPSS (version 22.0; IBM Corporation, Armonk, NY, USA) or MedCalc (version 15.2.2; MedCalc Software, Mariakerke, Belgium). All statistical tests were two-sided, and P < 0.05 was considered statistically significant.

Patient characteristics
A total of 151 patients who underwent CCTA before ICA and FFR were included. Nineteen patients were excluded due to an interval longer than 1 month between CCTA and ICA, and 8 patients were excluded because the corresponding vessels assessed with FFR were not feasible for CT-FFR or perivascular FAI measurement on CCTA imaging. Finally, 124 patients with 152 vessels were included for further analysis. The average age was 61.1 ± 9.2 years, and 64.5% were male. The prevalence of hypertension, diabetes mellitus, hyperlipidemia, current smoking and family history were 64.5%, 34.7%, 42.7%, 30.6%, and 17.7%, respectively. The medication history of patients is also shown in Table 1.

Vessel characteristic distribution was different according to severity of calcification
At the per vessel level, 152 vessels were included for measurement of coronary artery morphology ( Table 2). According to the FFR, 47 vessels were evaluated as hemodynamically significant and 105 vessels have none hemodynamically abnormality. All the measured characteristics except CACS/vessel and ≥ CADRADS 4 proportion, were significantly different between the FFR ≤ 0.80 and FFR > 0.80 groups. The perivascular FAI in the FFR ≤ 0.80 group was significantly higher than that in the FFR > 0.80 group (-73.0 ± 7.5 vs. -79.0 ± 7.4 HU, P < 0.001). The into the multilayer recursive neural network to optimize the sequence model. Lesion-specific CT-FFR was defined as the simulated FFR value at a distal point located 10 ~ 20 mm of the target lesion. For diffuse lesions (length ≥ 30 mm) or tandem lesions with continuous interval ≤ 10 mm, the measurement location is 10~20 mm away from the end of the lesion. For discontinuous tandem lesions with interval greater than 10 mm, it is recommended to measure the CT-FFR value at the 10~20 mm distal away from the most distant lesion.

ICA and FFR detection
ICA was performed according to a standard protocol either via the femoral or radial approach, and ICA imaging was independently evaluated by two experienced investigators with > 15 years' experience who were blinded to the patients' clinical and CCTA data. FFR was performed clinically to evaluate the necessity of revascularization. FFR was performed in all major vessel coronary branches unless their lumen diameter was either < 30% or > 90% stenosis. FFR was measured using a pressure wire (PressureWire Certus, St. Jude Medical), which was introduced through a 6 or 7 F catheter into the coronary artery. Intracoronary adenosine triphosphate was injected manually through the median cubital vein at 140 ~ 180 µg/(kg·min) for hyperemia. The FFR pressure wire was positioned a minimum of 20 mm distal to the stenosis in vessel segments ≥ 2 mm. The lowest recorded stable FFR value during hyperemic steady state was registered. FFR value ≤ 0.80 was considered indicative of lesion-specific hemodynamic coronary stenosis.

Statistical analysis
Data were analyzed either at the pre patient or per vessel level. Continuous data were presented as the mean ± standard deviation or median with quartiles. Comparisons between two groups were assessed by Student's t test for normally distributed data or Mann-Whitney U test for nonnormally distributed data. Categorical data were presented as percentages and were compared using the chi-square test or Fisher's exact test according to the cell size. The correlation analysis was performed using Pearson or Spearman rank correlation. Inter-observer agreement of perivascular FAI and CT-FFR was assessed among 50 randomly selected vessels independently analyzed by 2 observers. Spearman rank correlation coefficient was used for this analysis. For evaluation of diagnostic performance, the area under the curve (AUC) with 95% confidence interval (CI) was further calculated based on receiver operating characteristic curve (ROC). The optimal cutoff value was determined by the maximum sum of sensitivity and specificity, and the accuracy, sensitivity, specificity, positive predictive value The diagnostic efficacy of CT-FFR was impaired in severely calcified vessels Using a CT-FFR value of 0.80 as the cutoff value, the sensitivity, specificity and accuracy of the CT-FFR of each calcified group were calculated. The results are shown in Fig. 2. A significant decrease in specificity (P = 0.033) and accuracy (P = 0.010) was noted as calcification increased, whereas no significant difference was observed in sensitivity between the two stratifications. Furthermore, the specificity of CT-FFR diagnosis decreased significantly in severely cal-

Perivascular FAI provides incremental diagnostic value for identifying hemodynamically abnormality in severe calcified vessels
To demonstrate the relationship between CT-FFR or perivascular FAI versus FFR, we analyzed the correlation of different calcified severity groups, as shown in Fig. 3. A good positive correlation was noted between CT-FFR and FFR, both in the nonsevere and severe calcification groups (0.68 and 0.59). A negative correlation was also observed between perivascular FAI and FFR in different calcified groups (r=-0.22 and − 0.38, respectively).
Because CT-FFR and perivascular FAI demonstrated significant differences between different FFR groups in the univariable analysis, the diagnostic test was used to show the diagnostic performance of these two indices. The AUC, accuracy, sensitivity, specificity, PPV and NPV are shown in Table 3; Fig. 4. In nonsevere calcification, CT-FFR showed correlation of intra-observer and inter-observer measurements were both good for perivascular FAI (correlation coefficient: intra-observer: 0.96; inter-observer: 0.97).
Based on the severity of calcification, we compared the vascular parameters in different FFR groups. In the nonsevere calcified stratification, CACS/vessel, lesion length, ≥CADRADS 4 proportion were not significant, and CT-FFR was lower in vessels with FFR ≤ 0.80 than that in vessels with FFR > 0.80. The perivascular FAI between the two groups was significantly different (-73.2 ± 7.5 vs. -78.2 ± 7.4 HU, P = 0.002). Moreover, in the severe calcified stratification, no significant difference in ≥ CADRADS 4 proportion was noted between the different FFR groups, but the perivascular FAI of the FFR ≤ 0.80 group was significantly higher than that of the FFR > 0.80 (-72.8 ± 7.7 vs. -82.7 ± 6.3 HU, P < 0.001). The CT-FFR in vessels with FFR ≤ 0.80 was still lower than that in vessels with FFR > 0.80 (0.80 ± 0.06 vs. 0.84 ± 0.04, P = 0.012).

Discussion
This study investigated the impact of calcification on CT-FFR in the diagnosis of hemodynamically significant CAD and explored the incremental diagnostic value of perivascular FAI over CT-FFR. Compared with nonsevere calcified   large-scale registry studies [6][7][8]. Researchers have also performed some investigations to explore the accuracy of CT-FFR diagnosis under different scenarios, such as calcification [20][21][22]. However, whether severe calcification affects the diagnostic accuracy of CT-FFR remains controversial. In a sub-study of the NXT study, Norgaard et al. concluded that the diagnostic power of CT-FFR in severe calcification was not significantly different from that in mild calcification vessels (AUC: 0.91 vs. 0.95, P = 0.65) [21]. In vessels, the diagnostic performance of CT-FFR for the diagnosis of hemodynamic CAD was significantly impaired by severely calcification. Furthermore, in combination with CT-FFR and perivascular FAI, the diagnostic performance in predicting hemodynamically significant CAD could be greatly improved (AUC: 0.740, 95% CI: 0.573-0.869 vs. 0.919, 95% CI: 0.784-0.983, P = 0.036).
The value of CT-FFR in the diagnosis of hemodynamically significant CAD has been verified in several the CRISP CT study demonstrated that perivascular FAI can predict all-cause death and cardiac death [12].
Local perivascular fat is associated with the development and progression of CAD as well as a reduction in the FFR value. Previous studies have confirmed that perivascular FAI is a sign of perivascular fat edema, which is caused by the release of inflammatory factors from coronary plaques and circulation [10]. Vascular inflammation is an important cause of endothelial dysfunction. If coronary plaque is associated with impaired diastolic capacity of the stenotic segment, a decrease in pressure and a positive FFR will occur under adenosine-induced hyperemia. On the other hand, perivascular fat inflammation also promotes the occurrence of vulnerable plaque, and there are local hemodynamic changes in vulnerable plaque region, which may lead to reduced FFR [23]. In addition, the subsequent myocardial ischemia caused by lesion specific ischemia also causes some inflammatory changes in the myocardium, which may have an effect on the adjacent perivascular fat. Therefore, this is reason why a clinical correlation was noted between perivascular FAI and hemodynamic disorders.
The combination of perivascular FAI with total plaque volume and maximum diameter stenosis rate has high diagnostic accuracy for hemodynamic CAD [11]. These results suggest that as an extravascular imaging index, perivascular FAI can be used in the diagnosis evaluation of hemodynamic CAD. Actually, Brandt et al. paid attention to the ability of epicardial fat volume in confirming hemodynamics. They revealed that incremental discriminatory value with the addition of epicardial fat volume to plaque measures alone (AUC 0.84 vs. 0.62) [24]. Compared with whole epicardial fat, morbid perivascular fat may have a greater effect on pathophysiological changes in the coronary artery, contrast, Tesche et al. revealed the diagnostic performance of CT-FFR in different degrees of calcification and found that CT-FFR had better diagnostic ability compared with CCTA. However, when the CACS was greater than 400, the diagnostic performance of CT-FFR decreased significantly (AUC: 0.71 vs. 0.85, P = 0.04) [22]. The current study was generally consistent with the above Tesche's results, and the following reasons may explain the results: (1) The manual drawing of extravascular contours may be affected by calcification, resulting in overestimation of lesion stenosis and an eventual overestimation of CT-FFR. This process reduces the specificity of measuring CT-FFR in severe calcified vessels, as shown in Fig. 6. (2) Another reason may be the fact that more vessels with a negative FFR (105 vessels) were included in our study, which could potentially lead to a decrease in specificity. (3) In addition, our study also differed from previous studies in the stratification of calcification severity.
In recent years, perivascular FAI has attracted more attention as a new imaging biomarker associated with CAD and cardiac prognostic risk [10]. Perivascular fat may release a large number of inflammatory factors into the vascular endothelium, resulting in local vascular inflammation. Vascular inflammation can also subsequently affect perivascular fat, leading to lipid transformation into the aqueous phase. A dual effect is noted between coronary vessels and perivascular fat. This effect is reflected in an increase in fat CT attenuation. Therefore, the change in CT attenuation related to vascular inflammation can serve as a detectable indicator for CAD risk stratification. Antonopoulos et al. reported a difference in perivascular FAI between CAD patients and healthy subjects as well as between patients with acute myocardial infarction and stable angina pectoris [10]. Moreover, Although the CT-FFR result showed a positive lesion result (CT-FFR 0.80), coronary stenosis was only 70% when the ICA was performed, and FFR was negative (FFR 0.93) so diagnosis combined with perivascular FAI could be a promising approach for hemodynamics ischemia.
However, perivascular FAI alone remains limited in predicting hemodynamic CAD. This finding may be attributed to the fact that plaque development was in the early stage in the nonsevere calcification group with slight calcification and slight vessel inflammatory activity. Surprisingly, in patients with severe calcification, the sensitivity of perivascular FAI in predicting hemodynamic CAD increased to 72.2%, which may be associated with the high proportion of FFR positivity in these patients. For diffuse calcification related to active vascular inflammatory reaction, perivascular FAI was more likely to demonstrate a high value aided in the recognition of more positive cases in patients with severe calcification. Therefore, it is desirable to measure the CT-FFR value but also to refer to the information of perivascular FAI for a comprehensive evaluation [25][26][27]. The present study extends earlier findings by demonstrating the high and superior diagnostic performance of CT-FFR combined with perivascular FAI in a wide range of coronary calcification severities. The findings in this study support the potential of combination indices as a reliable gatekeeper to ICA across a representative cohort of patients.
We have to admit several limitations in our study. First, this was a retrospective analysis with a small sample size, in which the number of positive samples was relatively small. Because all the patients underwent ICA and FFR examination, most of the patients were at high risk for CAD, and differences may be noted in the general population of lowmiddle risk individuals. In addition, we applied CT-FFR technology based on ML. CT-FFR derived on this algorithm was still limited application in clinical practice. Third, part of the volume effect may extend to the outside of the adventitia in vessels with extreme calcification, resulting in an increase in local perivascular FAI. Caution should be taken when perivascular FAI is applied to vessels with CACS greater than 300. Furthermore, the inclusion number of relative more negative invasive FFR vessels and the different calcification severity stratification from other studies may reduce the overall diagnostic efficacy of CT-FFR.

Conclusion
For coronary artery with severe calcification, the diagnostic performance of CT-FFR in discriminating flowlimiting lesions could be greatly impaired. Perivascular FAI represents a potential reliable imaging marker to provide incremental diagnostic value over CT-FFR for identifying hemodynamically significant ischemia with severe calcification.