Data of this single-center study were derived from ongoing PET-MRI research projects. Thirty patients with suspected CMD were prospectively enrolled between September 2024 and March 2025 and examined at the first affiliated hospital of University of Science and Technology (USTC). The selection of CCTA or invasive coronary angiography was determined by cardiologists based on patient symptoms and coronary risk factors. Inclusion criteria for CMD included: (1) symptoms of myocardial ischemia and (2) non-obstructive coronary arteries (defined as <50% stenosis diameter). Exclusion criteria included: (1) presence of myocardial scar on late gadolinium enhancement (LGE), (2) history of coronary revascularization (including percutaneous coronary intervention or coronary artery bypass graft), (3) acute myocardial infarction, (4) atrial fibrillation, (5) severe valvular heart disease, (6) heart failure, (7) cardiomyopathy, (8) chronic kidney disease (eGFR < 45 mL/min), and (9) contraindications to intravenous regadenoson or CMR. All participants were specifically advised to avoid caffeine for 24 h before imaging. Ethics approval was obtained from the Ethics Committee of The First Affiliated Hospital of USTC of China for study procedures, and written informed consent was obtained from all participants prior to participation.

All participants underwent an overnight fast of at least 6 h, except for water intake. PET-MRI scans were acquired on a 3.0 T PET-MRI system (Biograph mMR, Siemens Healthineers, Erlangen, Germany). Each patient underwent simultaneous ¹³N-ammonia PET and gadoterate dimeglumine (Gd-DTPA) perfusion MRI scans at rest and during regadenoson-induced hyperemia.

All patients performed the rest imaging first, followed by stress imaging approximately 1 h later. For rest PET-MRI imaging, a 10 min dynamic PET perfusion scan during rest was started simultaneously with the administration of 370–555 MBq of ¹³N-ammonia (using a second intravenous cannula) followed by a bolus of saline (20 mL at 3 mL/s). MRI perfusion imaging was conducted concurrently with the PET scan. After the start of the PET scan, a single bolus of Gd-DTPA contrast agent (Beilu Pharmaceutical, Beijing, China) (0.075 mmol/kg body weight) was administered via a power injector at a flow rate of 4 mL/s into an antecubital vein, immediately followed by a 20 mL normal saline flush. A saturation recovery turboFLASH CMR sequence was used for MRI perfusion imaging using electrocardiography (ECG)-triggered and breath holding technique. Key parameters included: TR/TE = 2.67/1.15 ms, flip angle of 20°, field of view (FOV) = 380 × 304 mm, slice thickness = 8 mm, matrix size 128 × 102, and a pre-pulse delay of 100 msec.

Stress imaging employed identical parameters to the rest perfusion sequence described above, except that regadenoson was administered 60–90 s prior to simultaneous intravenous injections of ¹³N-ammonia and Gd-DTPA. Regadenoson (400 μg) was administered via an antecubital cannula as a single intravenous bolus over less than 10 s, immediately followed by 5 mL saline flush. Dynamic PET images were reconstructed using a three-dimensional ordered-subset estimation-maximization (3D-OSEM) with 6 iterations and a 5 mm post-reconstruction filter. The dynamic frame protocol included 12 × 10 s, 2 × 30 s, 4 × 60 s and 1 × 180 s intervals using the two-point Dixon attenuation correction.

Left ventricular myocardial perfusion analysis was performed using the standardized 16-segment model defined by the American Heart Association (15). Myocardial blood flow was quantified for both the entire left ventricle and three myocardial regions corresponding to the coronary artery territories (LAD, left anterior descending; LCX, left circumflex artery; and RCA, right coronary artery). CFR was calculated as the stress-to-rest MBF ratio and analyzed at both global and regional levels, respectively.

Image analysis was independently performed by two reviewers. If there was a disagreement between the two, a third observer joined the evaluation. The final result was determined by consensus among all three. Interobserver agreement for image analysis was assessed using kappa coefficient for categorical variables and intraclass correlation coefficient (ICC) for continuous variables. A kappa value >0.75 and ICC > 0.80 were considered to indicate excellent agreement. Regions of interest were defined in the MRI images using CVI42 post-processing software (version 5.1.1, Calgary, Canada). Endocardial and epicardial contours of the left ventricle were manually traced to derive rest and stress data, with subsequent generation of time-signal intensity curves for myocardial tissue and cardiac blood pool. Concordance between CMR and PET was evaluated on a per-vessel basis.

All PET data were quantified semi-automatically using a commercially available dedicated software package (Syngo MBF®, Siemens Healthcare, Erlangen, Germany), generating MBF-values for the entire left ventricle and in three regions corresponding to the coronary artery territories. Mean CFR was calculated as the ratio of stress MBF (stress scan) to rest MBF (rest scan) in each region and globally.

Continuous variables were expressed as mean ± standard deviation (SD) or median with interquartile range (IQR). Categorical variables were expressed as frequency with percentage. Pearson’s correlation was used to quantify association between continuous variables. Bland–Altman analysis was performed to evaluate the agreement between CMR and PET. Diagnostic performance of CMR indices for detecting CMD was evaluated through receiver-operating characteristic (ROC) curve analysis. Statistical analyses were conducted using the IBM SPSS Statistics v26.0 (SPSS Inc., Chicago, IL), GraphPad Prism v9.0 (GraphPad Software, La Jolla CA), and MedCalc v19.1 (MedCalc Software, Ostend, Belgium). For all analyses, a two-tailed p-value < 0.05 was considered significant.

Baseline characteristics and hemodynamic parameters of the cohort of 30 patients (11 were classified as CMD and 19 as the reference group with preserved CFR) are summarized in Tables 1, 2, respectively. The protocol for the study is illustrated in Figure 1.

Although sex, BMI, and age did not differ significantly between the two study groups, patients with impaired CFR tended to be older and had higher BMI values compared to those with preserved CFR. The study group had a higher incidence of diabetes, hypertension, and hyperlipidemia than the control subjects.

Figure 2 shows the relationship between MRI and PET measurements of global myocardial perfusion. On a per patient basis, MRI-derived CFR (CFR_MRI) and PET-derived CFR (CFR_PET) showed moderate correlation (r = 0.64; p < 0.001) and moderate inter-method reliability (ICC for absolute agreement = 0.66 [95% confidence interval (CI): 0.28 to 0.84]; p = 0.003). Bland–Altman analysis demonstrated a mean bias of − 0.79 ± 0.85 for CFR. Table 2 (the second to last row) displays the mean values of global CFR as measured using MRI and PET. Global CFR was significantly lower for MRI in comparison to PET (1.77 ± 0.49 vs. 2.56 ± 1.08; p < 0.001). MRI-derived MBF (MBF_MRI) and PET-derived MBF (MBF_PET) obtained during rest and stress showed correlation (r = 0.50, p = 0.005 and r = 0.49, p = 0.006, respectively). Bland–Altman analysis demonstrated a mean bias of −0.83 ± 0.47 mL/g/min and −1.84 ± 0.57 mL/g/min for rest and stress MBF, respectively.

The relationship between MRI and PET measurements of regional myocardial perfusion is illustrated in Figure 3. On a per vessel basis, correlation (r = 0.48; p < 0.001) and inter-method reliability (ICC for absolute agreement = 0.57 [95% CI: 0.34 to 0.71]; p < 0.001) were present between CFR_MRI and CFR_PET. Bland–Altman analysis revealed a mean bias of − 0.74 ± 0.98 for CFR. MRI demonstrated a tendency to underestimate CFR at both patient and vessel levels. Table 2 (the bottom row) lists the mean values of MRI and PET measurements of CFR. CFR_MRI was significantly lower than CFR_PET (1.81 ± 0.62 vs. 2.55 ± 1.09; p < 0.001). MBF_MRI and MBF_PET obtained during rest and stress showed only poor to moderate correlation (r = 0.26, p = 0.015 and r = 0.43, p = 0.019, respectively). Bland–Altman analysis showed a mean bias of − 0.85 ± 0.48 mL/g/min and − 1.89 ± 0.64 mL/g/min for rest and stress MBF, respectively.

Figure 4 shows the ROC curves of MRI perfusion imaging for detecting impaired CFR as defined by ¹³N-ammonia PET. CFR_MRI displayed an area under the curve (AUC) of 0.847 (95% CI: 0.669 to 0.952) and an optimal cutoff value of 1.57 (sensitivity: 72.73%; specificity: 89.47%). Figure 5 shows a 39-year-old symptomatic male in the impaired CFR group. First-pass MRI (Panel A) reveals uniform rest perfusion but diffuse mid-apical signal reduction under stress, consistent with microvascular dysfunction. Corresponding ¹³N-ammonia PET (Panel B) shows matching stress tracer uptake defects. Quantitative analysis (Panels C–E) demonstrates moderate agreement in global CFR (MRI: 1.20; PET: 1.48), aligning with the cohort’s overall correlation (r = 0.64), while stress MBF differs more notably (MRI: 1.12 vs. PET: 2.12 mL/g/min), reflecting the broader trend of larger MBF discrepancies. This case exemplifies the study’s key findings: consistent CFR for CMD diagnosis despite variable absolute MBF.

To further explore the impact of comorbidities, and given that diabetes was the only condition associated with a significant CFR difference (Table 3), we conducted subgroup analyses in patients with and without diabetes (Figures 6, 7). Despite the known microvascular alterations in diabetes, the overall correlation between MRI and PET for CFR persisted, though myocardial blood flow (MBF) measurements exhibited greater variability in diabetic patients—aligning with characteristics of diabetic microvascular dysfunction. Importantly, the diagnostic performance of MRI-derived CFR (using the established cutoff of 1.57 for CMD) remained robust across both subgroups. These results indicate that while diabetes influences microvascular perfusion (evident in MBF variability), it does not negate the core agreement between MRI and PET for evaluating global CFR, thus supporting the primary findings.

First, the study enrolled relatively small sample size. Nevertheless, this is the first study in patients and larger than previous volunteer-based studies. Second, respiratory control differences—MRI requiring breath-holds versus PET allowing free breathing—may introduce variability in MBF quantification. MRI breath-holds minimize motion artifacts, improving spatial resolution for regional MBF analysis, but brief breath-holding could transiently alter hemodynamics (e.g., subtle heart rate changes) that affect stress-induced hyperemia. In contrast, PET’s free-breathing approach avoids such transient physiological shifts but may introduce motion-related blurring, particularly in inferior myocardial segments, potentially diluting regional MBF accuracy. These differences likely contribute to the observed discrepancies in MBF between modalities, as motion artifacts and physiological perturbations disproportionately affect absolute perfusion values. Third, while PET MBF and CFR analysis are performed robustly and automatically within minutes using commercial software, MRI analysis remains time-consuming, operator-dependent, and error-prone. Finally, CFR interpretation should be exercised with caution when assigning myocardial segments to coronary artery territories based on common vascular distribution patterns. However, interindividual variability in coronary anatomy may render such assumptions potentially inaccurate.