The main findings of this study demonstrated that DP modeling outperformed the standard Fermi modeling in the setting of obstructive CAD detection, in both per vessel and per patient analysis. When compared with visual and quantitative CMR analysis, DP modeling-derived MBF at stress in per patient analysis, showed the highest diagnostic performance for the detection of obstructive CAD in our pilot population.
Visual versus quantitative CMR analysis
The interobserver variability for visual analysis was similar to a previously published value [
10] and higher compared to another study [
9]. In per vessel based analysis, visual CMR estimates gave higher specificity compared to quantitative CMR estimates (Tables
2 versus 6). However, quantitative CMR analysis showed superior sensitivity using both Fermi and DP modeling-derived MBF at stress, compared to visual CMR assessments.
In per patient based analysis, DP modeling demonstrated superior diagnostic performance in detecting obstructive CAD, compared to visual CMR estimates (Tables
2 versus 6).
Distributed parameter versus Fermi analysis
This is the first study assessing the diagnostic performance of 1-barrier 2-region DP modeling in 24 patients with known or suspected CAD, against invasive methods. Studies assessing the diagnostic performance of fully quantitative perfusion CMR methods in patients have mainly focused on the use of Fermi modeling [
5,
9,
10,
19], or on a model-independent approach [
20], for detecting obstructive CAD. A recent perfusion CMR study has assessed four different model applications (Fermi model, uptake model, one-compartmental model, model-independent approach) at 1.5T, in which it was shown that the diagnostic performance of quantitative MR analysis did not significantly differ between modeling methods, for obstructive CAD detection [
11]. In our data analysis, the diagnostic power of Fermi modeling was in agreement with previously published studies [
5,
9‐
11,
19]. However, DP modeling showed significantly higher diagnostic performance compared to Fermi modeling, which was consistent throughout both per vessel and per patient based analysis.
To date, no other perfusion CMR study has accurately identified perfusion abnormalities in the presence of significant CAD using only MBF value at stress as a measure (estimated per epicardial vessel territory). Perfusion CMR studies have instead defined the use of the lowest scoring segments for detecting obstructive CAD. This approach was applied either by using MPR as measure [
5,
10,
19], or MBF at stress as measure [
20], or both [
11]. Patel et al. used Fermi modeling-derived MPR, to assess quantitative MR analysis in patients with obstructive CAD [
9]. In our data, no differences were observed between all three haemodynamic parameters for either model, other than the case of the Fermi modeling derived-MPR
2, which showed improvement against MBF at stress and MPR, in per vessel analysis. This agrees with previous studies demonstrating no difference between MBF at stress and MPR when the measure of the lowest scoring segments was examined [
11,
20].
The optimal thresholds in per vessel (1.75 mL/min/mL) and in per patient (2.00 mL/min/mL) analysis for DP modeling-derived MBF at stress (Table
4), were in agreement with a previous positron emission tomography (PET) myocardial perfusion study (1.85 mL/min/mL) which aimed to localise perfusion defects to significantly stenotic coronary arteries in per vessel and per patient basis (≥70 % on invasive angiograms) [
21]. It is important to note that PET is currently considered the reference standard for absolute non-invasive quantification of MBF [
1,
5,
21].
The Fermi model demonstrated higher haemodynamic thresholds in both per vessel (2.49 mL/min/mL) and per patient (2.60 mL/min/mL) based analysis, compared to DP modeling (Table
4). Threshold values for Fermi modeling were in agreement with a previously published value (2.30 mL/min/mL) [
22]. However, thresholds for Fermi modeling were in a slightly higher range compared to other PET perfusion studies (1.86, 2.50 and 2.45 mL/min/mL) [
23‐
25] respectively, which aimed to identify perfusion abnormalities against lower angiographic thresholds (≥50 % on invasive angiograms). In our analysis, higher perfusion estimates were observed for Fermi modeling compared to DP modeling, across all three haemodynamic parameters (Table
3, Bland Altman plot analysis). In the same context, other perfusion studies demonstrated that the Fermi model estimates MBF values that were systematically increased compared to DP modeling [
12,
26], to two-compartmental modeling, model-independent and Patlak model analysis [
27], as well as against PET imaging data analysed with the Patlak model [
5]. It is known that arterial input function saturation effects that may be present in single bolus data can result in significant overestimation of MBF in Fermi modeling [
28]. Our group has previously demonstrated that the DP model can be less dependent on arterial input function saturation effects, compared to the Fermi model [
12]. Any MBF overestimations using Fermi modeling may have become pronounced in some of our subjects due to saturation effects in our single bolus data. This may explain its lower sensitivity in detecting hypoperfusion in obstructive CAD (susceptible to false negatives), compared to DP modeling.
Per vessel versus per patient quantitative analysis
The significant differences between (no, minor or non)-obstructive and obstructive CAD (Table
4 and
5), the AUC (Table
4, Figs.
3 and
5) and the diagnostic performance (Table
6) of both models were considerably superior in per patient analysis, compared to per vessel analysis. DP modeling-derived MBF at stress in per patient analysis demonstrated the highest diagnostic performance in detecting impaired haemodynamics in obstructive CAD (Table
6) and compares favourably against previous findings [
9,
11]. These outcomes indicate that it may have merit for the stratification of patients with at least one vessel with obstructive CAD.
The specificities and positive predictive values in per vessel based analysis were in agreement with those reported by a previous study [
19], but in a lower range compared to other investigations [
5,
10]. Quantitative CMR analysis identified hypoperfusion in vessels with (no, minor or non)-obstructive CAD, which increased the false positives in per vessel analysis. 72 % of the patients with (no, minor or non)-obstructive disease had at least one vessel with obstructive CAD (Table
1). Also, the vast majority of the study participants had been referred for angina (96 %), were under treatment for cardiac arrhythmias and hypertension (beta blockers, 83 %), were under medication for hyperlipidemia (statins, 88 %) and were previous smokers (63 %), all high risk factors for microvascular dysfunction [
29]. It is important to consider that microvascular dysfunction may have a major impact on global MBF [
6,
29], which may also have affected myocardial perfusion in vessels with (no, minor or non)-obstructive CAD. The coincidental effect of microvascular dysfunction in these patients, could possibly explain the homogeneous deficiency in coronary blood flow detected with both quantitative modeling approaches.
Study limitations
The main limitation of this study is the small population size. However, this is a pilot study to assess the feasibility of applying the DP model in this cohort of patients with known or suspected CAD. The above methods need to be assessed in larger patient cohorts to further assess their diagnostic accuracy. For perfusion CMR, a single bolus protocol was implemented to eliminate patient discomfort, similar to previous quantitative perfusion CMR studies at 1.5T [
11,
19,
20] and at 3T [
10]. Thus, it was impossible to assess any MBF overestimations at the specific contrast agent dose (0.05 mmol/kg) used in this study, due to arterial input function saturation issues at 3T. Our group has previously shown that the DP model was less dependent on saturation effects, although at a lower contrast agent dose (0.03 mmol/kg) [
12]. However, it is currently shown here that DP modeling achieved higher sensitivity and specificity in detecting obstructive CAD in our pilot cohort. This suggests that further investigation is required to determine whether DP modeling may be a more robust method of analysis for single bolus data at 3T, compared to the Fermi model. Any possible misregistration between the actual architecture of vessel territories and the standard 16-segment model used for myocardial segmentation [
14] is a methodological consideration that should not be excluded in both visual and quantitative CMR analysis. Both types of analysis can be subject to overlap of vessel territories which could in turn affect their sensitivity and/or specificity [
21]. Despite this, the reference method for quantitative CMR analysis of myocardial perfusion still occurs across the three major epicardial arteries [
14] and this standard type of analysis was also implemented in this study.