Analysis of bi-atrial function using CMR feature tracking and long-axis shortening approaches in patients with diastolic dysfunction and atrial fibrillation

Sixty healthy controls and ninety cardiovascular disease patients with either heart failure [15], paroxysmal or persistent AF, and/or obstructive coronary artery disease were included in the analysis (Supplemental Fig. 1). From clinical echocardiography exams performed within 28 days of the CMR exam, the grade of diastolic dysfunction was acquired from the patients’ clinical diagnosis written in the report and further verified with diastolic dysfunction guidelines based on the reported echocardiographic measurements[3]. Patients were then grouped into one of four categories: no diastolic dysfunction (grade 0), diastolic dysfunction grade I, a combined group of diastolic dysfunction grade II or III, or as having AF. Healthy controls were defined as subjects 18 to 70 years of age without cardiovascular or respiratory disease or medication that would impact these systems. All participants had provided written consent for secondary use of data. This protocol was reviewed and approved by the ethics committee of the canton of Bern. (MACDAVD study, approval number 2020_01258).

Images were obtained with a 3.0-Tesla clinical scanner (Siemens Prisma or Skyra, Siemens Healthineers). Retrospective-gated cine images with 25 − 30 phases were acquired in short-axis and long-axis views orientated to the left ventricle. In the case of varying RR intervals, gating was changed to prospective triggering at the discretion of the imaging technician. T1 maps (5(3)3-modified look-locker inversion recovery) were acquired in three short-axis views. In cardiovascular patients, 0.1–0.15 mmol/kg of contrast agent (Gadovist™, Bayer AG), was administered and T1 maps were reacquired post-contrast for the assessment of extracellular volume (ECV) while late gadolinium enhancement (LGE) images were acquired for assessment of focal scar[16].

All images underwent a second level of coding to blind the reader from the originating study and site, disease status, and other variables. Analysis was performed with cvi⁴² (version 5.12–5.13, Circle Cardiovascular Imaging). Ventricular parameters were measured as previously described [16, 17].

Detailed participant characteristics are provided in Supplemental Table 1. Of the 90 patients, 24 (27%) patients were classified with diastolic dysfunction grade I, 20 (22%) patients with diastolic dysfunction grade II/III, and 20 (22%) patients had atrial fibrillation (paroxysmal n = 12, persistent n = 8). The remainder had normal diastolic dysfunction or were not classified. Heart failure was diagnosed in n = 53 (59%) of the total cohort and was frequent in the diastolic dysfunction grade I (n = 17, 71%), grade II/III (n = 18, 90%), and AF subgroups (n = 13, 65%). Heart failure with preserved ejection fraction (HFpEF) was more common than heart failure with reduced ejection fraction (HFrEF) in both diastolic dysfunction grade I (HFpEF = 71% vs HFrEF = 29%) and diastolic dysfunction grade II/III (HFpEF = 94% vs HFrEF = 6%). The opposite was seen in patients with AF with a higher prevalence of HFrEF (HFpEF = 37% vs HFrEF = 69%). None of the patients with normal diastolic function had heart failure (n = 0, 0%).

The left ventricular analysis could be acquired from all 150 datasets. Concerning the LA, nine patients were excluded from the analysis because 2CH and 4CH views were off the plane with foreshortening, in which both the atrial roof and the middle of the valve were not visible. Thus, 94% of the cines were available for analysis. All remaining images could be analyzed for volumetric and rapid LAS atrial analysis, with 84% eventually being usable for FT analysis after excluding another seven for tracking errors. Similarly for the RA, sixteen images were excluded due to plane position, thus 89% were usable for volumetric and atrial LAS measurements which could be performed on all appropriate planes, with 81% eventually being usable for the FT strain analysis (Fig. 2, Supplemental Fig. 1).

Correlograms in Fig. 3 demonstrate that the highest linear correlation coefficients were found between FT strain and LAS, while associations between FT strain rate and atrial volumetry were weaker. In particular, when focusing on the reservoir (peak) strain of the LA, a good correlation of r = 0.83, p < 0.01 was observed between the FT and LAS measurements for all participants. For this comparison, analysis from the Bland-Altman yielded a non-significant bias of  − 3.0 ± 9.3% for the LAS measurements in comparison to FT (Supplemental Fig. 2, Supplemental Table 2). Even when performing this analysis for just patients with diastolic dysfunction (r = 0.75, p < 0.01) or AF (r = 0.88, p < 0.01), similar comparisons were observed between the two techniques (Supplemental Table 3). For the same comparison in the RA, a correlation of r = 0.66, p < 0.01 was observed with a bias of − 0.2 ± 12.5% when investigating all participants, with equal comparisons observed in the subsets of patients with diastolic dysfunction (r = 0.65, p < 0.01) and AF (r = 0.73, p < 0.01). Significant, but weaker linear relationships were observed for the conduit and booster atrial phases.

In comparison to the older control subgroup (FT: 43 ± 12%, LAS: 37 ± 9%), the patient subgroups with diastolic dysfunction (grade I: 30 ± 10% and 30 ± 8%, grade II/III: 24 ± 13% and 22 ± 11%, for FT and LAS respectively) or AF (FT: 14 ± 11% and LAS: 15 ± 11%) had a significantly reduced reservoir strain (p < 0.05). On the other hand, significant reductions in LA ejection fraction only occurred with the grade II/III diastolic dysfunction and AF subgroups (Fig. 4, Supplemental Table 4). Consequently, logistic regression showed that only FT and LAS measurements could discriminate patients with diastolic dysfunction grade I from older controls, while all techniques were able to discriminate the later grades of diastolic dysfunction and AF (Fig. 5). Reduced strain measurements in the conduit and booster phase were observed with diastolic dysfunction grade II/III by both FT and LAS. Patients with AF also showed a reduced conduit function, and no booster function was detected (Fig. 6).

With RA analysis, no significant findings were observed for either the RA reservoir or RA passive measurements in grade 0 or grade I patients in comparison to the older control group. Only diastolic dysfunction grade II/III and AF subgroups demonstrated attenuated atrial function measurements for these two phases (Supplemental Table 5). Similar to the LA, LAS and FT in the RA were able to discriminate patients in earlier subgroups than RA ejection fraction (Fig. 5).

In comparison to the traditional ventricular measurements of ejection fraction which only showed attenuation in the AF subgroup, ventricular markers of LV-LAS, MAPSE, and TAPSE from the LAS and valve tracking module demonstrated attenuation with the diastolic dysfunction grades I, II/III, and AF subgroups (Table 1). LV tissue characterization mirrored these results, showing that native T1 in grade 0 was similar to the controls, grade I, II/III, and AF subgroups all had higher native T1.

These findings indicate that both FT-strain and LAS measures may allow for detecting atrial abnormalities at an early stage of disease, consistent with echocardiography investigations [1, 18, 19]. Speckle tracking techniques have shown that LA strain is abnormal when left atrial volume indices are within reference ranges in patients with diastolic dysfunction [20], and an independent publication reported that replacing left atrial volume index with atrial strain reduced the proportion of indeterminate cases of diastolic dysfunction [21].

The key advantage of the FT and LAS techniques in comparison to volumetry is that they can be analyzed using minimal planes. Mono-planar and bi-planar volumetric analysis is a calculated representation of the atrial volume based on lumen diameters and area assuming an ideal spheric geometry. For more accurate volumetric analyses, full-stack atrial imaging would be required. However, when using current segmented cine techniques these additional slices required for full coverage volume analysis add time to image acquisition and analysis; consequently, these new FT strain and LAS approaches offer a comprehensive analysis from a relatively parsimonious amount of data. In addition, these new approaches may be more sensitive than using biplanar volumetric assessment. Echocardiography studies have shown that LA peak strain is less load-dependent than atrial volumetry-derived function measurements [22]. However, 2D imaging still relies on the quality and location of image acquisition, and especially for the atria, acquisition axes are generally defined based on LV anatomy (Fig. 2).

Interestingly, although our RA results show the same trends as in the LA data, significant findings were typically observed only with more severe diastolic dysfunction or AF. For instance, attenuated RA reservoir FT strain and LAS were not observed until diastolic dysfunction grade II/III was present, and no technique could discriminate diastolic dysfunction grade I from the RA. Pathophysiological explanations may be based on backward failure from progressive left-heart disease, or direct interventricular interactions. Since we categorized our groups by grading LV diastolic dysfunction, it appears logical that increased LV diastolic pressures first induce LA enlargement and dysfunction. Over time, this can cause a backward failure chain reaction with pulmonary venous congestion, eventually exposing the RV and RA to increased filling pressures. Direct left-to-right interactions may occur since all heart chambers are confined in the same non-distensible pericardium. Here, the advantage of interrogating all four chambers is that the function of each atrial and ventricular chamber can be assessed individually, also during identical phases of respiratory cycling, providing insight into the individual’s pathway of heart failure development. For example, in a recent pulmonary hypertension study, different atrial-ventricular interactions were observed in patients with pulmonary arterial hypertension secondary to HFpEF when compared with patients without HFpEF [23]. It is important to evaluate both sides of the heart because RA enlargement compared to LA size is an independent predictor for mortality [24]. A simultaneous four-chamber investigation may thus indicate whether pathophysiology is driven by the left or right side. Although our investigation focuses on LV diastolic dysfunction, RV diastolic dysfunction should not be discounted.

Atrial measurements are influenced by the dynamics of the atrio-ventricular interaction. Atrial reservoir (peak) function is dependent on both atrial stiffness and ventricular contractility during systole. On the other hand, conduit function is dependent on both atrial contractility and early LV relaxation. The ventricle is likely to play a larger role in atrial function in the early development of diastolic dysfunction. As the ventricle with diastolic dysfunction gradually stiffens, the atrium distends owing to increased atrial pressure, which in turn maintains blood flow into the ventricle in early diastole. We observed the earliest atrial FT-strain and LAS dysfunction during the LA reservoir phase (diastolic dysfunction grade I). This is consistent with previous findings from an echocardiography study where LA reservoir strain showed better agreement with invasively measured filling pressures, a hallmark of diastolic dysfunction, than the echocardiography marker E/e’ [25].

The LAS analysis module applied in our analysis is also useful because it provides results for ventricular systolic function, i.e. TAPSE, MAPSE, and LV-LAS, which were analyzed simultaneously with the atria. In comparison to the LV ejection fraction, which was only abnormal in the AF group, ventricular LAS along with the valve tracking markers of MAPSE and TAPSE detected abnormal ventricular dysfunction as early as diastolic dysfunction grade I, in line with the atrial measures. A unique characteristic of CMR in comparison to echocardiography is that its findings can be linked with tissue characterization measurements, which also reflected increasing myocardial fibrosis with increasing grades of diastolic dysfunction indicating a stiff LV.

This comparative study of atrial function assessments using the FT and LAS techniques indicated similar output, especially for the reservoir phase. Because LAS analysis was not solely reliant on tracking algorithms and contours could be manually adjusted, more patients could be included in the LAS analysis when compared to FT assessment of the same images. Especially in the case of AF, all images could be assessed for LAS. FT requires a good visualization of the entire atrial walls throughout the cardiac cycle, which made it prone to more errors and a higher user-defined exclusion rate. As LAS analysis is a simple calculation of chamber shortening, this can theoretically be performed by manual analysis as well without advanced post-processing software and presents an accessible option for imaging sites that do not have access to the newest software. A topic of further study might be to investigate time expenditure for each analysis method. Future work should assess these analysis techniques in relation to clinical patient status and outcomes and determine their prognostic utility in a larger cohort.

Our classification of diastolic dysfunction relies on clinical echocardiographic reports performed within four weeks prior to the CMR exam. A change in hemodynamics, i.e., pre- and afterload, can impact diastolic dysfunction grading. A desirable approach in the future would be to conduct echocardiography and CMR on the same day and to compare CMR measurements directly to contemporary echocardiographic quantification of diastolic dysfunction, e.g., by transmitral blood flow and Doppler tissue velocity interrogation. Due to varying RR intervals, AF patients can have a different acquisition triggering scheme than the other subgroups. With prospective triggering late diastole is not imaged; however, in patients with AF, this should not impact measurements significantly as there should not be any change in chamber volume and size during this time due to a lack of atrial kick. Additionally, our sample size was small for the subgroup analysis. Further comparisons between software vendors are required due to differences in tracking algorithms and methodological approaches to contour placement.