Reproducibility of left atrial function using cardiac magnetic resonance imaging

Sixty participants were included: AS (n = 16), T2D (n = 28), end-stage renal disease on haemodialysis (n = 10) and healthy volunteers (n = 6). Participants were prospectively recruited for ethically approved studies at a single tertiary cardiac centre. Inclusion and exclusion criteria were as previously published: AS [20, 21], T2D [22, 23], and haemodialysis [24]. For the AS cohort, transthoracic echocardiography was performed on the same day to evaluate AS severity [25]. The healthy subjects [23] were non-diabetic, devoid of known cardiovascular disease and with normal electrocardiography (ECG), cardiopulmonary exercise testing, echocardiography and CMR.

All subjects underwent two CMR scans 7–14 days apart, on the same scanner for each participant, using standardised protocols. CMR scans for patients with AS (n = 10), T2D (n = 17), haemodialysis (n = 10) and healthy volunteers (n = 6) were conducted on a 3-Tesla scanner (Siemens Skyra). The rest of the subjects (n = 17) were scanned on a 1.5-Tesla scanner (Siemens Avanto or Siemens Aera). Haemodialysis patients were scanned on their non-dialysis day and not after their long break to standardise their volume status as far as possible. All subjects were in sinus rhythm and scanned using retrospective ECG gating. Long-axis (2-, 3- and 4-chamber) cine images were acquired before contrast administration in all subjects, using a steady-state free precession end-expiratory breath-hold sequence (typical parameters: slice thickness of 8 mm, matrix 256 × 204, field of view variable 300–360 × 360–420, TR 45 ms, TE 1.2 ms, flip angle 45°). Short-axis LV cine stacks were acquired to enable derivation of LV volumes, mass and function.

Image analysis was performed offline using dedicated software by a single trained observer (AA), blinded to subject details. Image quality was graded as 0 = not analysable, 1 = fair (artefact present but images still analysable), 2 = good (artefact present but not in the region of interest), 3 = excellent. Blinding was achieved by anonymising the scans using unique study codes, performing batch analysis of all scans in a random order and allowing at least 2 weeks’ gap for the intra-observer variability analysis.

LA maximal volume (LAVmax) at end-ventricular systole and LA minimal volume (LAVmin) at end-ventricular diastole were quantified using the biplane area length method [7] on 2- and 4-chamber cine images, using QMass v8.1 (Medis Suite v3.1 Medical imaging systems) (Fig. 1). LA volume pre-atrial contraction (LAVpre-A) was measured during ventricular diastole in the cine frame immediately prior to atrial contraction. LA area was automatically generated by the software after identifying the junction points of the anterior and posterior mitral annulus with the LA wall and an additional reference point on the LA roof (maximum length). Insufficient tracing was manually adjusted as required. The LA appendage and pulmonary veins were excluded from LA volumetric measurements. LAEF was calculated corresponding to the three LA phases: reservoir function (LA relaxation allowing the collection of venous return during LV systole), conduit function (passive LV filling during early diastole) and contraction booster-pump (active LV filling by LA contraction during late diastole [26] according to the following equations:

LAS and LASR were assessed with feature tracking, using QStrain v2.0 (Medis Suite v3.1, Medical imaging systems) from the 2- and 4-chamber cine images and average values calculated (Fig. 2). LA endocardial borders were traced at ventricular end-diastole and end-systole, excluding the LA appendage and pulmonary veins, and the software automatically propagated contours to the rest of the cardiac cycle. Contour adjustment was only possible on the end-ventricular diastole and end-ventricular systole phases, and was done in around 30% and 75% of the cases, respectively, to ensure the exclusion of the appendage and pulmonary veins from the LA volume. Strain and strain rate curves were obtained by identifying end-ventricular diastole as the time reference (a value of zero designated as the baseline) in line with prior recommendations [27]. LAS and LASR were also measured for the three LA phases [27]: reservoir function (LAS_r and LASR_r), conduit function (LAS_cd and LASR_cd) and contraction booster-pump (LAS_bp and LASR_bp). LAS_cd was calculated as: LAS_cd = LAS_r−LAS_bp.

Image analysis for test-retest reproducibility was conducted by a single observer (AA) who also performed the intra-observer variability assessment. The intra-observer assessment was performed on 10 scans that were randomly selected using an online random selection generator (a mixture of AS, T2D and healthy volunteers) for repeat analysis, performed at least 2 weeks apart. For inter-observer variability assessment, the same scans were analysed by a second blinded observer (BAM).

Statistical tests were performed using SPSS version 26.0 software (Statistical Package for the Social Sciences). Normality was assessed using the Shapiro-Wilk test, histograms and Q-Q plots. Numerical data are expressed as mean ± standard deviation (SD) and categorical data are expressed as counts and percentages. Test-retest reproducibility and intra/inter-observer variability were assessed using the Bland-Altman method to test the limits of agreement [28], two-way mixed-effect intraclass correlations (ICC) calculated for absolute agreement and the coefficient of variance (CoV) [29] to assess the range between the mean and SD of the difference. ICC was scored as follows: excellent > 0.90, good 0.75–0.90, moderate 0.50–0.75 and poor < 0.50 [30], while CoV was scored as excellent < 10%, good 10–20%, moderate 20–30% and poor > 30%. The differences were expressed as percentages in Bland-Altman plots, calculated as: [(scan1−scan 2)/mean) × 100] [31]. For continuous variables, one-way analysis of variance (ANOVA) was used to determine significant differences across the groups, whilst unpaired t tests were used to assess differences between two groups. Pearson’s correlation was used to assess correlation between techniques.

Most studies assessing reproducibility of an imaging technique focus on inter- and intra-observer variability [32‐34]. However, whilst important, observer variability does not address the multiple potential sources of variability when assessing the LA function, including image planning and acquisition, inclusion or exclusion of pulmonary veins/LA appendage from the LA volume as well as day-to-day physiological variation and filling status of the subject. Furthermore, studies assessing test-retest reproducibility of LA assessment are limited and with small sample sizes (n = 16–22) [35, 36] and none in clinical populations. Test-retest reproducibility of an imaging technique is fundamental for its validity and its appropriate use in longitudinal studies for monitoring changes with disease progression or in response to treatment.

We found LA volume and LAEF to be more reproducible than LA strain parameters. This has been noted previously in a study of 22 healthy subjects [35], although they used Tissue Tracking, a similar image analysis technique. Similar to our findings, LA strain corresponding to reservoir function has previously been shown to be the most reproducible strain parameter in 16 healthy subjects [36]. However, in that study, strain parameters were found to be more reproducible than volumes, contrary to our results. Their repeat scans were done on the same day, immediately following the first scan, which may have led to reduced variability in planning of the imaging, especially if the same radiographer performed both image acquisitions without blinding. The contradictory results may also be due to differences between various CMR image analysis platforms and vendor software packages, since the algorithm used to produce strain and strain rate curves differ [37]. Furthermore, especially for LA deformation analysis, the zero line for the strain curves is different across vendors [2, 27]. This is also true for LV deformation analysis where the test-retest reproducibility using tissue tracking has been found to be superior to that of feature tracking [38].

The poor test-test reproducibility of LA strain could also be related to the change LA filling states between the two scans. However, it has been shown that strain is less affected by preload than the volume assessment [39] and our date shows volumetric assessment to have better test-test reproducibility than strain.

Despite the calculation that is based on geometric assumptions [40], biplane area length has been increasingly used in clinical practice. This is due to it being a faster alternative to short-axis discs method, that does not require additional slices that increase the scan time, with more breathing instructions. It has been shown that the biplane area length method is a reliable and reproducible technique in CMR [41, 42]. This is supported by our results showing better test-retest reproducibility in comparison with LA strain analysis.

We found moderate correlations between the strain and volumetric parameters corresponding to the reservoir phase of LA function, which corresponds to the main LA assessment phase in routine clinical practice (total LAEF). One other previous study has looked at the correlation between the LA volume and LA strain parameters, and found good correlation, especially for reservoir function [33].

It has been shown that LA deformation has an incremental role in assessing disease progression and states, since it detects LA functional impairment at early stages before changes in LA size become evident [16, 43]. LA reservoir and conduit function using CMR feature tracking has recently been shown to be abnormal in the early stages of hypertension even before LVH develops [44]. We have shown that LAS and LASR assessments are feasible by CMR feature tracking using routinely acquired SSFP sequences, as previously reported [33]. Although tissue tagging is considered to be the gold standard for LV strain analysis by CMR, its utility in quantifying LA deformation is severely limited by spatial resolution, due to the LA wall being very thin [45]. CMR feature tracking therefore offers a feasible technique to quantify LA deformation, using routinely acquired cine images. However, it has poor test-retest reproducibility and is a relatively time-consuming technique in the clinical practice setting. Therefore, further developments are required before this technique can be recommended for use in routine clinical practice or as an outcome measure in clinical research studies.

This study has some limitations. The number of participants in each group is small and heterogeneous. However, test-retest reproducibility studies are rare, and the overall number is the largest to date reporting LAS reproducibility. The purpose was to assess the reproducibility of the imaging technique rather than assess differences between cohorts. We used a single software package for the image analysis and the result may not apply to other available software. Importantly, both strain and volume assessment were analysed using the same software. Patients were studied at both 1.5 Tesla and 3 Tesla scanners and hence some of the variability in our results may be due to inherent differences in reproducibility between field strengths. However, the two scans for each participant were performed on the same scanner with the same field strength. Blood pressure was not recorded during LA cine acquisition during the scans, and a baseline blood pressure was only recorded for the first CMR. However, medication remained unchanged between the two scans.