Abstract
Background
Cardiovascular magnetic resonance (CMR) accurately measures left ventricular (LV) ejection fraction (EF) and late gadolinium enhancement (LGE), but is not commonly used for diastolic function. This study combined unsupervised machine learning and diastolic function with cine imaging to identify prognostic groups.
Methods
CMR studies from 391 patients with LVEF ≥50%, and no LGE were included. Using LV cine diastolic time-volume curves, early peak filling rate (E-PFR) indexed to stroke volume index (SVI), early to late diastolic filling ratio (E/A) ratio, and deceleration time (Decelt) were measured, along with left atrial volume index (LAVI) and LV mass index (LVMI). K-means clustering grouped individuals using only these parameters. Primary outcome was a composite of heart failure or ventricular arrhythmia admission, heart transplant, mechanical support, and cardiac death.
Results
Median follow-up was 5 years with 30 events. Three clusters were identified with event rates of 2.6% (group 1), 8.5% (group 2), and 17.2% (group 3). Time-volume curves demonstrated a triphasic pattern with group 2 having lower E-PFR (6.5 vs 9.7 and 8.4 SVI/s for groups 1 and 3, respectively, p < 0.001), lower E/A (1.2 vs 2.4 and 1.9 for groups 1 and 3, respectively, p < 0.001), and longer Decelt (223 vs 79 and 105 ms for groups 1 and 3, respectively, p < 0.001). Group 3 had increased remodeling based on higher LAVI (51.4 vs 26.6 and 27.5 mL/m2 for groups 1 and 2, respectively, p < 0.001) and LVMI (78.7 vs 48.7 and 50.6 g/m2 for groups 1 and 2, respectively, p < 0.001). Compared to group 1, groups 2 and 3 had hazard ratios (HR) of 3.21 (95% confidence interval [CI] 1.06–9.68, p = 0.038) and 2.52 (95% CI 1.06–4.47, p = 0.002) for events, respectively. Adjusting for comorbidities, only group 3 had a significant HR of 2.55 (95% CI 1.42–4.57, p = 0.002) compared to group 1.
Conclusion
CMR cine-derived diastolic parameters can add prognostic value despite normal EF and absence of LGE.
Keywords: Cardiac magnetic resonance imaging, Diastolic dysfunction, Unsupervised learning, Cine imaging, Time-volume curves
Graphical abstract
1. Introduction
Assessment of diastolic function plays an important role in diagnosing heart failure and predicting adverse events. Echocardiography has been well-validated for estimating left ventricular (LV) filling pressure [1] and grading diastolic dysfunction [2]. In 347,408 patients with normal LV ejection fraction (EF) from the National Echo Database Australia [3], 9% had abnormal diastolic function by American Society of Echocardiography/European Association of Cardiovascular Imaging guidelines [4] and a significant increase in cardiovascular mortality. However, indeterminate diastolic function was reported for 22% of patients, demonstrating a common real-world limitation of assessment by echocardiography. Furthermore, studies have shown that patients with indeterminate diastolic function on echocardiography have a small but significant increase in risk for cardiovascular death [3].
Cardiovascular magnetic resonance (CMR) has become an invaluable tool in the diagnosis and management of cardiomyopathies. Advances in image processing and improvements in workflow, as well as recognition of its clinical utility, have led to a significant increase in the number of CMRs performed in the United States, with a three-fold increased use in the Medicare population over the last 10 years [5]. CMR can provide accurate volumetric measurements, perfusion, and tissue characterization in a single study [6]. Late gadolinium enhancement (LGE) imaging can identify myocardial fibrosis and predict adverse cardiovascular outcomes in non-ischemic cardiomyopathy [7]. Diastolic dysfunction can be detected with advanced CMR techniques, such as strain analysis, phase contrast imaging, and four-dimensional flow [8]. However, many of these methods for diastolic assessment can be technically challenging to implement in a clinical setting and, therefore, are not widely utilized.
Cine imaging is the most commonly used CMR technique and can provide important insight into diastolic function [9]. Rapid imaging approaches [10] and advanced segmentation tools [11], [12] have improved the accuracy of time-volume measurements with cine imaging. Early peak filling rate (E-PFR) and late peak filling rate (A-PFR) can be derived using the integral of CMR-derived LV time-volume curves to assess diastolic filling patterns, which correlate with diastolic dysfunction by echocardiography [13]. In addition, diastolic filling intervals such as time to peak filling rate or time to recover 80% of the stroke volume can be used to differentiate between grades of diastolic dysfunction [13]. Elevated left atrial (LA) volumes and LV mass are also CMR metrics associated with increased invasive pulmonary capillary wedge pressure (PCWP) and risk of death [14]. In this study, we assessed whether readily obtained diastolic parameters from cine imaging can identify distinct groups of patients based on parameters of diastolic function using unsupervised machine learning, and whether the individual groups are associated with a differential rate of adverse outcomes.
2. Methods
2.1. Study design
A retrospective multicenter registry of 1098 consecutive adult patients who underwent CMR for any clinical indication at University of Chicago Medicine and University of Virginia Health was used for this observational cohort study with approval by the institutional review board (Fig. 1). The clinical history and demographics were extracted from the electronic medical record. Patients were followed based on medical records or by telephone contact for clinical outcomes. Patients with significant artifacts or poor quality CMR images resulting in non-diagnostic interpretations were not included. Patients were excluded if they had CMR findings of LGE or reduced LVEF (<50%), pulmonary hypertension, congenital heart disease, cardiac amyloidosis, valvular heart disease (at least moderate), coronary artery bypass grafting (CABG), heart transplant, or history of prior left ventricular assist device (LVAD). CABG was excluded given that changes to the pericardial anatomy can affect diastolic filling. The primary outcome was major adverse cardiac events (MACE), defined as a composite of cardiac death, LVAD implantation, heart transplantation, and ventricular arrhythmia or heart failure admission.
Fig. 1.
Flowchart of patient recruitment. MRI magnetic resonance imaging
2.2. CMR protocol
CMR images were acquired with a 1.5T (Achieva, Philips, Best, the Netherlands or Aera, Siemens Healthineers, Erlangen, Germany) or 3T scanner (Skyra or Prisma, Siemens). Steady-state free precession sequence cine imaging was performed with retrospective gating during approximately 5–8 s breath holds (repetition time 2.9 ms, echo time 1.5 ms, flip angle 60°, temporal resolution 30–40 ms). Long-axis views were obtained in the two-, three-, and four-chamber. Short-axis cine images were acquired from the base to apex (slice thickness 8 mm, gap 2 mm, and 2.0 × 2.0 mm in-plane spatial resolution). The frame rate varied between 25 and 32 frames per cardiac cycle. Cine images were acquired either pre- or post-contrast. LGE imaging was performed in all subjects at least 10 min following administration of a gadolinium-based contrast agent in accordance with standard guidelines [6]. Images were acquired in short- and long-axis views using T1-weighted gradient echo with phase-sensitive inversion recovery (repetition time 4.5 ms, echo time 2.2 ms, inversion time 250–300 ms, flip angle 30°, voxel size 2 × 2 × 10 mm). Inversion time was selected using the Look‐Locker sequence to identify the optimal nulling of the normal myocardium [15].
2.3. Image analysis
The atria and ventricles were segmented with a semi-automated approach using an experienced operator (S.W.) SuiteHeart (Neosoft, Pewaukee, Wisconsin, USA). The maximum atrial volume was calculated based on the two- and four-chamber view at ventricular end-systole using modified Simpson’s method. For the right atrium, only the four-chamber was used. The left and right ventricular volumes and mass were calculated from the short axis using Simpson’s method of disks. For time-volume analysis, only the short-axis stack of the LV cavity volumes for each cardiac phase was collected. All chamber volumes and masses were indexed to body surface area (BSA), including LA volume index (LAVI) and LV mass index (LVMI). The absence of LGE was verified by independent readers (S.W. and J.P.) in all available views.
The LV time-volume measurements for the entire cardiac cycle were imported into R 4.3.1 (R Foundation for Statistical Computing, Vienna, Austria). A smoothing spline regression with restricted maximum likelihood was fitted to the LV time-volume measurements for the entire cardiac cycle and interpolated for 100 total phases. As a result, the average temporal resolution was 8.5 ms. LV time-filling rate curve was calculated as the difference in volume between phases divided by the time interval between phases throughout the cardiac cycle. The LV time-volume and time-filling curves were then truncated to only diastole, which was defined as the period between LV end-systole and end-diastole based on the smallest and largest LV volume, respectively. The diastolic LV time-filling rate curve profiles are similar to the mitral inflow Doppler profile on echocardiography, with an E-wave and an A-wave. From the LV time-filling curves, E-PFR and A-PFR were identified in an automated fashion based on the two largest upsloping peaks, with the earliest peak being E-PFR. In cases in which there was only 1 upsloping peak, both E-PFR and A-PFR were assigned the same value.
The LV time-volume and time-filling curves were used to measure the diastolic function parameters. These included indexed E-PFR, early to late diastolic filling ratio (E/A), time to peak early diastolic filling (E-PFRt), diastolic recovery time (Vol80%t), and early deceleration time (Decelt). Indexed E-PFR was calculated as E-PFR divided by LV stroke volume index (SVI). E/A ratio was determined as the ratio between E-PFR to A-PFR. From the time-filling curves, E-PFRt was the time from end-systole to E-PFR, which is equivalent to an early acceleration time. From the LV time-volume curves, Vol80%t was measured as the time from end-systole to when 80% of the LV stroke volume has recovered in diastole. Decelt was estimated by subtracting E-PFRt from Vol80%t, which is the time from early peak filling to when 80% of the LV stroke volume has recovered. In addition to LV time-volume and time-filling parameters, single-phase diastolic parameters were included, such as LVMI at end-diastole and LAVI at end-systole. CMR-estimated LV filling pressure was calculated with a previously validated equation [14]: PCWPCMR = 6.1352 + (0.07204 * LAvolume) + (0.02256*LVmass). Example segmentation, LV time-volume, and time-filling analysis are shown in Fig. 2.
Fig. 2.
Example of diastolic time-volume analysis. From short-axis cine images (A), the left ventricular cavity was segmented (red circle) at every cardiac phase using a semi-automated technique. Time-volume (B) and time-filling (C) curves during diastole were generated from the segmentations. Diastolic measurements such as early peak filling rate (E-PFR), atrial peak filling rate (A-PFR), time to fill 80% of the left ventricle (Vol80%), estimated deceleration time (Decelt), end-systolic volume (ESV), and end-diastolic volume (EDV) were collected in an automated fashion for each subject. LV left ventricular
2.4. Statistics analysis
Continuous variables were presented as mean and standard deviation, and categorical values were presented as numbers and percentages. All the statistical analyses were performed in R 4.3.1 (R Foundation for Statistical Computing, Vienna, Austria). Groups were compared with Wilcoxon rank-sum test or Kruskal-Wallis rank-sum test for continuous values and Fisher’s Exact test for categorical values. K-means clustering, a simple unsupervised learning approach, was used to group patients by diastolic CMR parameters. This approach was used to identify different CMR diastolic filling patterns, which can be highly variable and heterogeneous depending on the grade of diastolic dysfunction according to echocardiography literature [4]. The data were normalized by scaling using the mean and standard deviation. There was no missing data for diastolic cine parameters; therefore, imputation was not used. Diastolic parameters were selected for the model based on physiologic importance from echocardiography literature [4]. The optimal number of clusters was determined from the silhouette value. Kaplan-Meier curves were used for time-to-event analysis. Cox proportional hazard model was used to calculate hazard ratios (HR) with 95% confidence intervals (CI) for time-to-event data. P-values <0.05 were considered significant. Proportional hazard assumption was assessed for each variable using Schoenfield residuals. C-statistic and Wald Global chi-square were used to determine the incremental value of time-volume parameters.
3. Results
3.1. Patient demographics
A total of 391 patients were included, with a median follow-up time of 5.0 years and a total of 30 events. There were 19 heart failure admissions, 7 ventricular arrhythmia admissions, 4 cardiac deaths, and no heart transplants or LVAD implantations. The clinical characteristics and CMR measurements are shown in Table 1. The mean age was 51 ± 16 years and 59% were female. Patients who experienced MACE had significantly lower mean body mass index (BMI) (24 vs 29 kg/m2, p < 0.001). There was a significantly higher percentage of patients in the MACE group who had a family history of coronary artery disease (43 vs 22%, p = 0.013). There was no significant difference between those with and without MACE in age, sex, race, ethnicity, other comorbidities, or medications.
Table 1.
Clinical demographics.
| Characteristics | All subjects (n = 391) |
MACE− (n = 361) |
MACE+ (n = 30) |
p-value |
|---|---|---|---|---|
| Age (years) | 51 ± 16 | 51 ± 16 | 55 ± 14 | 0.171 |
| Male | 41 (160) | 42 (150) | 33 (10) | 0.703 |
| Body mass index (kg/m2) | 28 ± 7 | 29 ± 7 | 24 ± 5 | <0.001 |
| Race and ethnicity | ||||
| White | 54 (212) | 53 (192) | 67 (20) | 0.184 |
| Black | 32 (125) | 32 (116) | 30 (9) | 1.000 |
| Hispanic or Latino | 1 (2) | 2 (1) | 0 (0) | 1.000 |
| Asian | 5 (20) | 5 (19) | 3 (1) | 1.000 |
| History of tobacco use | 29 (114) | 32 (105) | 30 (9) | 0.681 |
| Family history of CAD | 24 (93) | 22 (80) | 43 (13) | 0.013 |
| CAD | 25 (100) | 23 (93) | 23 (7) | 1.000 |
| Hyperlipidemia | 34 (132) | 33 (119) | 43 (13) | 0.315 |
| Hypertension | 49 (191) | 48 (173) | 60 (18) | 0.255 |
| Diabetes | 17 (67) | 17 (62) | 17 (5) | 1.000 |
| Stroke | 4 (17) | 5 (17) | 0 (0) | 0.631 |
| Heart failure | 15 (57) | 14 (49) | 27 (8) | 0.061 |
| Medications | ||||
| Aspirin | 31 (120) | 30 (109) | 37 (11) | 0.685 |
| Beta blocker | 39 (152) | 39 (140) | 40 (12) | 1.000 |
| ACEI/ARB | 32 (125) | 31 (115) | 33 (10) | 0.842 |
| Statin | 41 (158) | 41 (146) | 40 (12) | 1.000 |
| Diuretics | 21 (83) | 20 (73) | 33 (10) | 0.104 |
Continuous variables are presented as mean ± standard deviation and categorical variables as percent (number)
Significance is defined as p-value < 0.05
ACEI angiotensin-converting enzyme inhibitors, ARB angiotensin II receptor blockers, CAD coronary artery disease
3.2. CMR parameters
The CMR measurements are shown in Table 2. There was no significant difference in ventricular function and size between groups. Subjects who experience MACE had significantly higher mean LVMI (65 vs 54 g/m2, p = 0.03). There was no significant difference in mean LAVI (p = 0.060), right atrial volume index (RAVI) (p = 0.158), and PCWPCMR (p = 0.158) between those with and without MACE. For the time-volume and time-filling parameters, subjects with MACE had significantly lower mean E-PFR (408 vs 341 mL/s, p = 0.02), indexed E-PFR (8.1 vs 6.7 SVI/s, p < 0.001), and E/A ratio (1.8 vs 1.5, p = 0.015). They also had significantly higher mean Vol80%t (324 vs 298 ms, p = 0.03). Univariate Cox proportional analysis is shown in Supplemental Table 1. Diastolic parameters associated with increased risk of MACE included E-PFR (HR 0.96 [95% CI 0.94–0.99] per 10 mL/s, p = 0.009), indexed E-PFR (HR 0.72 [95% CI 0.60–0.86] per 1 SVI/s, p < 0.001), and LVMI (HR 1.25 [95% CI 1.10–1.44] per 10 g/m2, p < 0.001). PCWPCMR did not reach statistical significance as a predictor of MACE (p = 0.083) in this cohort of individuals with preserved LVEF and no LGE.
Table 2.
Cine CMR parameters.
| CMR parameters | All subjects (n = 391) | MACE− (n = 361) | MACE+ (n = 30) | p-value |
|---|---|---|---|---|
| Heart rate (bpm) | 71 ± 13 | 71 ± 13 | 72 ± 12 | 0.630 |
| LVEDVI (mL/m2) | 83 ± 24 | 83 ± 24 | 83 ± 25 | 0.856 |
| LVESVI (mL/m2) | 33 ± 12 | 33 ± 12 | 33 ± 14 | 0.959 |
| LVSVI (mL/m2) | 50 ± 13 | 50 ± 14 | 50 ± 12 | 0.721 |
| LVEF (%) | 61 ± 6 | 61 ± 6 | 61 ± 7 | 0.856 |
| LVMI (g/m2) | 54 ± 17 | 54 ± 16 | 65 ± 31 | 0.026 |
| RVEDVI (mL/m2) | 88 ± 26 | 87 ± 25 | 96 ± 34 | 0.183 |
| RVESVI (mL/m2) | 36 ± 15 | 36 ± 14 | 41 ± 25 | 0.777 |
| RVSVI (mL/m2) | 50 ± 13 | 50 ± 14 | 50 ± 13 | 0.721 |
| RVEF (%) | 59 ± 8 | 59 ± 8 | 59 ± 12 | 0.761 |
| LAVI (mL/m2) | 31 ± 13 | 31 ± 14 | 36 ± 27 | 0.060 |
| RAVI (mL/m2) | 34 ± 15 | 34 ± 14 | 37 ± 21 | 0.719 |
| PCWP (mmHg) | 12.7 ± 2.3 | 12.7 ± 2.2 | 13.6 ± 3.2 | 0.158 |
| Time-volume/filling parameters | ||||
| E-PFR (mL/s) | 402 ± 157 | 408 ± 159 | 341 ± 129 | 0.021 |
| Indexed E-PFR (SVI/s) | 8.0 ± 2.3 | 8.1 ± 2.3 | 6.7 ± 1.8 | <0.001 |
| A-PFR (mL/s) | 269 ± 123 | 268 ± 124 | 279 ± 114 | 0.514 |
| E/A ratio (unitless) | 1.8 ± 1.0 | 1.8 ± 1.1 | 1.5 ± 1.1 | 0.015 |
| E-PFRt (ms) | 152 ± 34 | 151 ± 34 | 160 ± 32 | 0.137 |
| Vol80%t (ms) | 300 ± 107 | 298 ± 108 | 324 ± 95 | 0.030 |
| Decelt (ms) | 148 ± 96 | 146 ± 98 | 164 ± 81 | 0.113 |
Continuous variables are presented as mean ± standard deviation.
Significance is defined as p-value < 0.05.
A-PFR atrial peak filling rate, CMR cardiac magnetic resonance, Decelt estimated deceleration time, E/A early to late diastolic filling ratio, EDVI end-diastolic volume index, EF ejection fraction, E-PFR early peak filling rate, E-PFRt time to early peak filling rate, ESVI end-systolic volume index, LAVI left atrial volume index, LV left ventricular, LVMI left ventricular mass index, MACE major adverse cardiovascular events, PCWP pulmonary capillary wedge pressure, RAVI right atrial volume index, RV right ventricular, SVI stroke volume index, Vol80%t time to fill 80% of the left ventricle.
3.3. Comparison of diastolic clusters
Given the heterogeneity of diastolic filling patterns, unsupervised K-means analysis was used to identify clusters of patients using only five diastolic parameters (indexed E-PFR, E/A ratio, Decelt, LAVI, and LVMI). The optimal number of clusters was three, with 67.8% of the variance explained in the first two principal components (Fig. 3). Clusters were labeled as group 1 (n = 152), group 2 (n = 175), and group 3 (n = 64). CMR diastolic parameters for each group are shown in Fig. 4 and Supplemental Table 2. There was a triphasic pattern for the diastolic time-volume curve parameters (index E-PFR, E/A ratio, and Decelt). Group 2 had a significantly lower mean indexed E-PFR (6.5 vs 9.7 and 8.4 SVI/s for group 1 and 3 respectively, p < 0.001), lower mean E/A ratio (1.2 vs 2.4 and 1.9 for groups 1 and 3 respectively, p < 0.001), and longer mean Decelt (223 vs 79 and 105 ms for groups 1 and 3 respectively, p < 0.001) compared to the other groups. For the volumetric parameters, group 3 had a significant increase in mean LAVI (51.4 vs 26.6 and 27.5 mL/m2 for groups 1 and 2, respectively, p < 0.001) and mean LVMI (78.7 vs 48.7 and 50.6 g/m2 for groups 1 and 2, respectively, p < 0.001) compared to the other groups.
Fig. 3.
K-means cluster analysis of diastolic cine parameters. Unsupervised learning approach using K-means cluster analysis identified similar groups based on only cine CMR parameters. When clustering into 3 groups, the first two principal components explained 38.7% and 29.1% of variance, respectively (A). The optimal number of clusters was determined using the average silhouette method (B). CMR cardiac magnetic resonance
Fig. 4.
Boxplot of diastolic CMR parameters by K-means groups: boxplots for indexed E-PFR (A), E/A ratio (B), Decelt (C), LAVI (D), and LVMI (E) for K-means groups 1, 2, and 3. Group 2 had significantly lower indexed E-PFR and E/A ratio with a longer Decelt compared to the other groups. Group 3 had a significantly higher LAVI and LVMI compared to the other groups. P-value < 0.05 was considered significant (*). CMR cardiac magnetic resonance, Decelt estimated deceleration time, E/A early to late diastolic filling ratio, E-PFR early peak filling rate, LAVI left atrial volume index, LVMI left ventricular mass index
The clinical demographics for the groups can be seen in Table 3. Group 2 was significantly older (58 vs 46 and 47 years for groups 1 and 3, respectively, p < 0.001) and had a higher prevalence of diuretic use (27 vs 18% and 13% for groups 1 and 3, respectively, p = 0.019). Group 3 had significantly more black patients (53 vs 34% and 22% for groups 1 and 2, respectively, p < 0.001) and a higher prevalence of coronary artery disease (33% vs 18% and 18% for groups 1 and 2, respectively, p = 0.015). There was a significant difference in MACE rates between the groups (2.6% vs 8.5% vs 17.2% for groups 1, 2, and 3, respectively, p < 0.001).
Table 3.
K-means groups by diastolic cine parameters.
| Characteristics | Group 1 (n = 152) |
Group 2 (n = 175) |
Group 3 (n = 64) | p-value |
|---|---|---|---|---|
| Age (years) | 46 ± 17 | 58 ± 13 | 47 ± 16 | <0.001 |
| Male | 36 (55) | 42 (73) | 50 (32) | 0.165 |
| Body mass index (kg/m2) | 29 ± 8 | 28 ± 6 | 28 ± 6 | 0.830 |
| Race and ethnicity | ||||
| White | 50 (76) | 64 (112) | 38 (24) | <0.001 |
| Black | 34 (53) | 22 (38) | 53 (34) | <0.001 |
| Hispanic or Latino | 1 (1) | 0 (0) | 2 (1) | 0.154 |
| Asian | 5 (7) | 6 (11) | 3 (2) | 0.679 |
| History of tobacco use | 29 (44) | 30 (52) | 28 (18) | 0.977 |
| Family history of CAD | 20 (31) | 24 (42) | 31 (20) | 0.233 |
| CAD | 18 (27) | 18 (31) | 33 (21) | 0.015 |
| Hyperlipidemia | 34 (51) | 36 (63) | 28 (18) | 0.531 |
| Hypertension | 44 (67) | 54 (95) | 45 (29) | 0.155 |
| Diabetes | 18 (28) | 16 (28) | 17 (11) | 0.860 |
| Stroke | 4 (6) | 4 (7) | 6 (4) | 0.674 |
| Heart failure | 16 (24) | 12 (21) | 19 (12) | 0.340 |
| Medications | ||||
| Aspirin | 32 (48) | 33 (58) | 21 (14) | 0.233 |
| Beta blocker | 39 (59) | 41 (71) | 34 (22) | 0.710 |
| ACEI/ARB | 30 (46) | 35 (62) | 27 (17) | 0.374 |
| Statin | 41 (62) | 40 (70) | 41 (26) | 0.999 |
| Diuretics | 18 (27) | 27 (48) | 13 (8) | 0.019 |
| MACE (%) | 2.6 (4) | 8.5 (15) | 17.2 (11) | <0.001 |
Continuous variables are presented as mean ± standard deviation and categorical variables as percent (number)
Significance is defined as p-value < 0.05.
ACEI angiotensin-converting enzyme inhibitors, ARB angiotensin II receptor blockers, CAD coronary artery disease, MACE major adverse cardiovascular events
3.4. Prognostication by diastolic clusters
From the Kaplan-Meier estimate (Fig. 5), K-means groups were a significant predictor of MACE (p = 0.003). However, in this cohort, PCWPCMR > 15 mmHg was not a significant predictor of MACE (p = 0.99). Using group 1 as the reference, Cox proportional HR for MACE was 3.21 (95% CI 1.06–9.68, p = 0.038) for group 2 and 2.52 (95% CI 1.06–4.47, p = 0.002) for group 3. The HR between groups 2 and 3 was not significant (p = 0.10). After adjusting for age, group 2 was not a significant predictor of MACE compared to group 1 (p = 0.08). Group 3 continued to be a significant predictor of MACE after adjusting for age, CAD, and race, with a HR 2.55 (95% CI 1.42–4.57, p = 0.002) compared to group 1. Compared to CMR volumetrics alone (LAVI and LVMI), time-volume parameters (indexed E-PFR, E/A, and Decelt) increased the C-statistic discrimination from 0.60 to 0.67, but did not significantly improve the model based on Wald chi-square (p = 0.054).
Fig. 5.
Kaplan-Meier estimates of adverse events for K-means groups and estimated pulmonary capillary wedge pressure (PCWP). For K-means diastolic groups (A), there was a significant difference in adverse events based on the log-rank test (p = 0.003). However, patients with an estimated PCWP > 15 mmHg (B) did not have a significant increase in adverse events (p = 0.99). MACE major adverse cardiovascular events
4. Discussion
In this study, readily obtainable CMR parameters from cine imaging were used to predict the risk of adverse outcomes based on the severity of diastolic dysfunction. Patients with reduced EF or evidence of scar on LGE imaging were excluded, as studies suggest they already have an elevated risk of adverse events [7], [16]. We combined unsupervised learning with diastolic CMR parameters to identify phenotypic clusters and demonstrated significantly different levels of risk for MACE between the groups.
The findings of this study suggest that the diastolic CMR groups reflect different severities of diastolic dysfunction. Group 1 represents a relatively healthy cohort with a low rate of adverse events and no evidence of impaired relaxation or restrictive filling. Group 2 consists of subjects with impaired relaxation based on reduced indexed E-PFR, reduced E/A ratio, and prolonged Decelt. Group 2 was significantly older and likely included subjects with impaired relaxation associated with normal aging. In addition, many of the subjects in group 2 were being treated for heart failure symptoms based on the higher prevalence of diuretic use. Both of these groups also had noticeably smaller LAVI by CMR. Group 3 represents subjects with adverse remodeling and likely restrictive physiology. Group 3 had significantly elevated LAVI and LVMI, which is most often due to chronically elevated intracardiac pressures. The time-filling parameters of E-PFR, E/A ratio, and Decelt in group 3 were similar to those of group 1, which can be seen with pseudonormalization of the transmitral filling patterns. The unsupervised learning did not specifically cluster subjects with restrictive filling patterns and differentiate them from pseudonormal filling based on high indexed E-PPR or E/A, possibly due to low prevalence of severe diastolic dysfunction in this cohort.
In this study, risk stratification by clusters had similar findings to echocardiographic studies [3], [16]. In a study by Levene et al. [16] with 136,455 normal EF patients, severe diastolic dysfunction by the 2016 echocardiography guidelines was associated with a HR of 2.66 (95% CI 2.35–3.00) for cardiovascular hospitalizations. In our study, group 3 had a significant HR of 2.55 (95% CI 1.06–4.52, p = 0.004) for MACE compared to group 1, which was independent of age. On the other hand, PCWPCMR >15 mmHg was not a significant predictor of MACE (p = 0.99). This metric was validated by Garg et al. [14], where they enrolled 835 patients with suspected heart failure who underwent CMR, echocardiography, and invasive PCWP. However, this study did not exclude patients with LGE or reduced EF. In addition, they did not index LA volume or LV mass for BSA. In a separate study [17] evaluating PCWPCMR in patients with EF ≥50%, CMR had a lower area under the receiver operating characteristic curve (0.72 vs 0.87, p = 0.008) compared to echocardiography.
There are important technical differences between CMR and echocardiography that influence the characterization of diastolic dysfunction. E-wave velocity (cm/s) from Doppler echocardiography and volumetric flow (mL/s) from CMR are similar but are not the same. Mitral inflow velocity represents the velocity profile of blood at the leaflet tips, while the LV filling rate is the volumetric flow of blood across the mitral plane in diastole. Velocity encoded imaging with CMR can directly measure LV filling but requires careful slice selection and motion tracking. Cine imaging with LV segmentation indirectly measures LV filling, assuming no shunts or valvular regurgitation. These two methods correlate well with each other [18]. E-wave Decelt by CMR is feasible with sufficient temporal resolution [19]. As an alternative, cine imaging CMR can also measure diastolic volume recovery time, which is calculated as the time to recover 80% of the LV stroke volume [13]. Similar to Decelt by echocardiography, diastolic volume recovery time is associated with diastolic dysfunction [20]. Diastolic volume recovery time increases in grade I diastolic dysfunction and then decreases in grade II and III diastolic dysfunction [13]. In our study, we estimated Decelt as the difference between E-PFR and recovery of 80% of the stroke volume. Our estimated Decelt was relatively shorter (148 ± 96 ms) compared to velocity encoded CMR (187 ± 28 ms) and Doppler echocardiography (178 ± 43 ms) in normal healthy controls [19]. Finally, LV mass and LA volume measurements can vary significantly depending on modality. CMR is often considered the gold standard due to high contrast resolution, multiple imaging planes, and fewer geometric assumptions. Two-dimensional (2D) echocardiography estimates LV mass using linear wall measurements, which can be problematic due to the difficult differentiation of the LV wall and trabeculation, and the assumption that the LV is a prolate ellipsoid. On the other hand, CMR directly measures the myocardial volume in short axis over the entire LV to calculate the mass with better differentiation of the wall/trabeculation interface. In similar sample sizes, CMR can detect smaller changes in LV mass compared to 2D echocardiography [21]. While atrial volumes can be measured in short-axis cine imaging, they are commonly calculated using biplane modified Simpson’s method from the two- and four-chamber views, similar to 2D echocardiography. Nevertheless, 2D echocardiography frequently underestimates LA volumes compared to CMR and 3D echocardiography [22]. Therefore, LV mass and LA volumes between modalities should not be used interchangeably. In our study, increased LVMI and LAVI by CMR identified those at highest risk for cardiovascular events and therefore may warrant closer follow-up.
From these diastolic groups, we propose an algorithm for classifying risk using only diastolic cine CMR parameters in individuals with preserved LVEF and no LGE (Fig. 6). The proposed algorithm consists of three steps to grade diastolic dysfunction. The optimal cut-offs were derived using Youden’s index to differentiate between the diastolic groups identified in our study. In the first step, LAVI and LVMI by CMR are used to screen patients for signs of high-grade diastolic dysfunction and chronically elevated filling pressures. Pathologic LV hypertrophy has been shown to contribute to increased myocardial stiffness and increased filling pressure [4], [23]. Similarly, LA dilation reflects adverse remodeling due to atrial hypertension and diastolic dysfunction [4], [24]. Furthermore, an increase in LA volume and LV mass by CMR is associated with a 1.8-fold increased risk of death [14]. In the second step, a prolonged Decelt and decreased E/A ratio differentiates patients with impaired relaxation and/or pseudonormal filling pattern. Patients with slowed LV relaxation have decreases in E-wave amplitude with either a smaller decrease or substantial increase in A-wave amplitude, which has been studied in elderly patients [25]. Decelt of the E-wave is a function of both myocardial relaxation and stiffness [26]. As a result, patients with primarily mild diastolic dysfunction with delayed relaxation will have prolonged E-wave Decelt, while those with severe restrictive physiology will have short Decelt [27]. For the final step of the algorithm, low-indexed E-PFR identifies early evidence of reduced diastolic filling. Early diastolic filling is primarily determined by myocardial relaxation and LA pressure [28]. When LA pressure is elevated from severe diastolic dysfunction, there is an increase in E-wave velocity [29]. However, in the early stages of diastolic dysfunction in which LA pressure is normal, slow myocardial relaxation results in a reduction of LA-LV pressure gradient and a reduced peak filling rate [30]. In CMR studies [20], [31] using short-axis cine imaging, there is predominantly a decrease in E-PFR indexed to stroke volume in patients with diastolic dysfunction, especially grade I by echocardiography.
Fig. 6.
Proposed algorithm for grading diastolic dysfunction. The algorithm was designed based on differences between the diastolic CMR groups. For step 1, elevation in left atrial volume index (LAVI) and left ventricular mass index (LVMI) is used to identify those with a high likelihood for restrictive filling pattern. For step 2, deceleration time (Decelt) and E/A ratio can differentiate patients with likely pseudonormal filling and/or impaired relaxation. For step 3, a low-indexed early peak filling rate (E-PFR) is used to identify those who are beginning to develop impaired relaxation. MACE major adverse cardiovascular events, CMR cardiac magnetic resonance, E/A early to late diastolic filling ratio
5. Limitations
This was a two-center retrospective cohort study. Patients referred for CMR may have a different pretest probability for diastolic dysfunction compared to those referred for echocardiography and invasive hemodynamic testing. There were likely unmeasured confounders that contributed to referral bias. Patients scheduled for CMR were not routinely screened for diastolic dysfunction based on echocardiography or invasive testing—thus a direct comparison is not available. Parameters such as clinical symptoms, New York Heart Association class, presence of atrial fibrillation, conditions associated with lower BMIs, specific heart failure medications such sodium-glucose cotransporter 2 inhibitors and mineralocorticoid receptor antagonists, or prior history of ventricular arrhythmia or sudden cardiac death were not inquired about at the time of CMR and were not retrospectively extracted from medical records. Because of the sample size, we were limited to a few covariates for multivariable analysis and were unable to include a validation cohort. Time-volume and time-filling parameters were measured in an automated fashion, which could result in minor sampling errors. Strain was not assessed in this cohort and will need to be evaluated in subsequent studies. Further investigation will need to be conducted to assess the generalizability of our results, especially compared to echocardiography.
6. Conclusion
CMR diastolic parameters from cine imaging alone can predict adverse cardiovascular events in patients with normal LVEF and no LGE. They can identify physiologically and prognostically meaningful clusters of patients that closely match validated grades of diastolic dysfunction. Leveraging the strengths of CMR, a simple algorithm for grading diastolic dysfunction using cine imaging is proposed. However, future validation and the incorporation of novel CMR techniques are needed to improve the assessment of diastolic function and impact clinical care.
Funding
This study was partially funded through the generosity of Mr. Para Suriyaarachchi. Jonathan Pan was supported by the Cardiovascular Imaging training grant 5T32EB003841.
Author contributions
Sivam Bhatt: Data curation. Roberto M. Lang: Supervision, Conceptualization. Christopher M. Kramer: Writing – review & editing, Supervision, Funding acquisition. Amit R. Patel: Writing – review & editing, Supervision, Resources, Methodology, Investigation, Conceptualization. Alper Turgot: Formal analysis. Cristiane De Carvalho Singulane: Methodology, Data curation. Deyu Sun: Software, Data curation. Yu Wang: Software, Data curation. Jonathan A. Pan: Writing – review & editing, Writing – original draft, Software, Methodology, Formal analysis, Conceptualization. Shuo Wang: Writing – review & editing, Methodology, Data curation, Conceptualization.
Declaration of competing interests
The author is an Editorial Board Member/Editor-in-Chief/Associate Editor/Guest Editor for this journal and was not involved in the editorial review or the decision to publish this article. The authors declare the following financial interests/personal relationships which may be considered as potential competing interests: Christopher M. Kramer—Eli Lilly - research support, consultant. Amit R. Patel—Unrelated research grants from GEHealthcare, Area19, NIH R01. Research support from Neosoft, CircleCVI, and Siemens Healthineers. Ownership in HeartMotion, LLC. Associate Editor for JCMR. Jonathan A. Pan—Supported by 5T32EB003841.
Footnotes
Supplementary data associated with this article can be found in the online version at doi:10.1016/j.jocmr.2026.102737.
Appendix A. Supplementary material
Supplementary material
.
References
- 1.Sugimoto T., Dohi K., Tanabe M., Watanabe K., Sugiura E., Nakamori S., et al. Echocardiographic estimation of pulmonary capillary wedge pressure using the combination of diastolic annular and mitral inflow velocities. J Echocardiogr. 2013;11(1):1–8. doi: 10.1007/s12574-012-0142-0. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2.Redfield M.M., Jacobsen S.J., Burnett J.C., Jr., Mahoney D.W., Bailey K.R., Rodeheffer R.J. Burden of systolic and diastolic ventricular dysfunction in the community: appreciating the scope of the heart failure epidemic. JAMA. 2003;289(2):194–202. doi: 10.1001/jama.289.2.194. [DOI] [PubMed] [Google Scholar]
- 3.Playford D., Strange G., Celermajer D.S., Evans G., Scalia G.M., Stewart S., et al. Diastolic dysfunction and mortality in 436 360 men and women: the National Echo Database Australia (NEDA) Eur Heart J Cardiovasc Imaging. 2021;22(5):505–515. doi: 10.1093/ehjci/jeaa253. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4.Nagueh S.F., Smiseth O.A., Appleton C.P., Byrd B.F., 3rd, Dokainish H., Edvardsen T., et al. Recommendations for the evaluation of left ventricular diastolic function by echocardiography: an update from the American Society of Echocardiography and the European Association of Cardiovascular Imaging. J Am Soc Echocardiogr. 2016;29(4):277–314. doi: 10.1016/j.echo.2016.01.011. [DOI] [PubMed] [Google Scholar]
- 5.Reeves R.A., Halpern E.J., Rao V.M. Cardiac imaging trends from 2010 to 2019 in the Medicare Population. Radiol Cardiothorac Imaging. 2021;3(5) doi: 10.1148/ryct.2021210156. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6.Kramer C.M., Barkhausen J., Bucciarelli-Ducci C., Flamm S.D., Kim R.J., Nagel E. Standardized cardiovascular magnetic resonance imaging (CMR) protocols: 2020 update. J Cardiovasc Magn Reson. 2020;22(1):17. doi: 10.1186/s12968-020-00607-1. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7.Kuruvilla S., Adenaw N., Katwal A.B., Lipinski M.J., Kramer C.M., Salerno M. Late gadolinium enhancement on cardiac magnetic resonance predicts adverse cardiovascular outcomes in nonischemic cardiomyopathy: a systematic review and meta-analysis. Circ Cardiovasc Imaging. 2014;7(2):250–258. doi: 10.1161/CIRCIMAGING.113.001144. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8.Ipek R., Holland J., Cramer M., Rider O. CMR to characterize myocardial structure and function in heart failure with preserved left ventricular ejection fraction. Eur Heart J - Cardiovasc Imaging. 2024 doi: 10.1093/ehjci/jeae224. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9.Chamsi-Pasha M.A., Zhan Y., Debs D., Shah D.J. CMR in the evaluation of diastolic dysfunction and phenotyping of HFpEF: current role and future perspectives. JACC Cardiovasc Imaging. 2020;13(1 Pt 2):283–296. doi: 10.1016/j.jcmg.2019.02.031. [DOI] [PubMed] [Google Scholar]
- 10.Tolouee A., Alirezaie J., Babyn P. Nonrigid motion compensation in compressed sensing reconstruction of cardiac cine MRI. Magn Reson Imaging. 2018;46:114–120. doi: 10.1016/j.mri.2017.11.004. [DOI] [PubMed] [Google Scholar]
- 11.Wang S., Patel H., Miller T., Ameyaw K., Narang A., Chauhan D., et al. AI based CMR assessment of biventricular function: clinical significance of intervendor variability and measurement errors. JACC Cardiovasc Imaging. 2022;15(3):413–427. doi: 10.1016/j.jcmg.2021.08.011. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12.Goyal N., Mor-Avi V., Volpato V., Narang A., Wang S., Salerno M., et al. Machine learning based quantification of ejection and filling parameters by fully automated dynamic measurement of left ventricular volumes from cardiac magnetic resonance images. Magn Reson Imaging. 2020;67:28–32. doi: 10.1016/j.mri.2019.12.004. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13.Mendoza D.D., Codella N.C., Wang Y., Prince M.R., Sethi S., Manoushagian S.J., et al. Impact of diastolic dysfunction severity on global left ventricular volumetric filling - assessment by automated segmentation of routine cine cardiovascular magnetic resonance. J Cardiovasc Magn Reson. 2010;12(1):46. doi: 10.1186/1532-429X-12-46. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14.Garg P., Gosling R., Swoboda P., Jones R., Rothman A., Wild J.M., et al. Cardiac magnetic resonance identifies raised left ventricular filling pressure: prognostic implications. Eur Heart J. 2022;43(26):2511–2522. doi: 10.1093/eurheartj/ehac207. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15.Messroghli D.R., Radjenovic A., Kozerke S., Higgins D.M., Sivananthan M.U., Ridgway J.P. Modified Look-Locker inversion recovery (MOLLI) for high-resolution T1 mapping of the heart. Magn Reson Med. 2004;52(1):141–146. doi: 10.1002/mrm.20110. [DOI] [PubMed] [Google Scholar]
- 16.Levene J., Voigt A., Thoma F., Mulukutla S., Bhonsale A., Kancharla K., et al. Patient outcomes by ventricular systolic and diastolic function. J Am Heart Assoc. 2024;13(4) doi: 10.1161/JAHA.123.033211. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17.Rahi W., Hussain I., Quinones M.A., Zoghbi W.A., Shah D.J., Nagueh S.F. Noninvasive prediction of pulmonary capillary wedge pressure in patients with normal left ventricular ejection fraction: comparison of cardiac magnetic resonance with comprehensive echocardiography. J Am Soc Echocardiogr. 2024;37(5):486–494. doi: 10.1016/j.echo.2024.02.001. [DOI] [PubMed] [Google Scholar]
- 18.Komi S., Inoue Y., Hata H., Nakajima A., Miyatake H. Cardiovascular magnetic resonance evaluation of left ventricular peak filling rate using steady-state free precession and phase contrast sequences. Springerplus. 2016;5(1):1163. doi: 10.1186/s40064-016-2878-x. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19.Cirener L.C., Körperich H., Barth P., Racolta A., Piran M., Burchert W., et al. Assessing diastolic function using CMR as an alternative to echocardiography: age- and gender-related normal reference values. Clin Res Cardiol. 2024;114(2):1641–1657. doi: 10.1007/s00392-024-02553-9. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20.Kawaji K., Codella N.C., Prince M.R., Chu C.W., Shakoor A., LaBounty T.M., et al. Automated segmentation of routine clinical cardiac magnetic resonance imaging for assessment of left ventricular diastolic dysfunction. Circ Cardiovasc Imaging. 2009;2(6):476–484. doi: 10.1161/CIRCIMAGING.109.879304. [DOI] [PubMed] [Google Scholar]
- 21.Myerson S.G., Bellenger N.G., Pennell D.J. Assessment of left ventricular mass by cardiovascular magnetic resonance. Hypertension. 2002;39(3):750–755. doi: 10.1161/hy0302.104674. [DOI] [PubMed] [Google Scholar]
- 22.Perez de Isla L., Feltes G., Moreno J., Martinez W., Saltijeral A., de Agustin J.A., et al. Quantification of left atrial volumes using three-dimensional wall motion tracking echocardiographic technology: comparison with cardiac magnetic resonance. Eur Heart J - Cardiovasc Imaging. 2014;15(7):793–799. doi: 10.1093/ehjci/jeu001. [DOI] [PubMed] [Google Scholar]
- 23.Hieda M., Sarma S., Hearon C.M., Dias K.A., Martinez J., Samels M., et al. Increased myocardial stiffness in patients with high-risk left ventricular hypertrophy. Circulation. 2020;141(2):115–123. doi: 10.1161/CIRCULATIONAHA.119.040332. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 24.Teo S.G., Yang H., Chai P., Yeo T.C. Impact of left ventricular diastolic dysfunction on left atrial volume and function: a volumetric analysis. Eur J Echocardiogr. 2009;11(1):38–43. doi: 10.1093/ejechocard/jep153. [DOI] [PubMed] [Google Scholar]
- 25.Bryg R.J., Williams G.A., Labovitz A.J. Effect of aging on left ventricular diastolic filling in normal subjects. Am J Cardiol. 1987;59(9):971–974. doi: 10.1016/0002-9149(87)91136-2. [DOI] [PubMed] [Google Scholar]
- 26.Shmuylovich L., Kovács S.J. E-wave deceleration time may not provide an accurate determination of LV chamber stiffness if LV relaxation/viscoelasticity is unknown. Am J Physiol Heart Circ Physiol. 2007;292(6):H2712–H2720. doi: 10.1152/ajpheart.01068.2006. [DOI] [PubMed] [Google Scholar]
- 27.Oh J.K., Park S.-J., Nagueh S.F. Established and novel clinical applications of diastolic function assessment by echocardiography. Circ Cardiovasc Imaging. 2011;4(4):444–455. doi: 10.1161/CIRCIMAGING.110.961623. [DOI] [PubMed] [Google Scholar]
- 28.Thomas J.D., Choong C.Y., Flachskampf F.A., Weyman A.E. Analysis of the early transmitral Doppler velocity curve: effect of primary physiologic changes and compensatory preload adjustment. J Am Coll Cardiol. 1990;16(3):644–655. doi: 10.1016/0735-1097(90)90356-t. [DOI] [PubMed] [Google Scholar]
- 29.Kuwaki H., Takeuchi M., Chien-Chia Wu V., Otani K., Nagata Y., Hayashi A., et al. Redefining diastolic dysfunction grading: combination of E/A ≤0.75 and deceleration time >140 ms and E/ε' ≥10. JACC Cardiovasc Imaging. 2014;7(8):749–758. doi: 10.1016/j.jcmg.2014.05.002. [DOI] [PubMed] [Google Scholar]
- 30.Ohno M., Cheng C.P., Little W.C. Mechanism of altered patterns of left ventricular filling during the development of congestive heart failure. Circulation. 1994;89(5):2241–2250. doi: 10.1161/01.cir.89.5.2241. [DOI] [PubMed] [Google Scholar]
- 31.Aquaro G.D., Pizzino F., Terrizzi A., Carerj S., Khandheria B.K., Di Bella G. Diastolic dysfunction evaluated by cardiac magnetic resonance: the value of the combined assessment of atrial and ventricular function. Eur Radiol. 2019;29(3):1555–1564. doi: 10.1007/s00330-018-5571-3. [DOI] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Supplementary material