Normal ranges for automatic measurements of tissue Doppler indices of mitral annular motion by echocardiography. Data from the HUNT3 Study

Automatic quantification of left ventricular (LV) function could enhance workflow for cardiologists and assist inexperienced clinicians who perform focused cardiac ultrasound. We have developed an algorithm for automatic measurements of the mitral annular plane systolic excursion (MAPSE) and peak velocities in systole (S′) and early (e′) and late (a′) diastole. We aimed to establish normal reference values for the automatic measurements and to compare them with manual measurements.


| INTRODUC TI ON
Automatic analysis of echocardiograms is a rapidly evolving technology. Potential advantages include, but are not limited to, timesavings for cardiologists, reduced measurement variability, and assistance for less experienced users of both echocardiography and focused cardiac ultrasound.
We have developed an algorithm that automatically measures indices of left ventricular (LV) longitudinal function, namely the mitral annular plane systolic excursion (MAPSE) and the peak mitral annular velocities in systole (S′) and early (e′) and late (a′) diastole. The algorithm uses color tissue Doppler imaging (cTDI) echocardiographic recordings from high-end scanners. 1 All the indices have shown prognostic value in heart failure. [2][3][4][5] The automatic measurements have previously been shown to be in good agreement with measurements by experts 1 and could be used to detect LV dysfunction. 6 The algorithm can be implemented on hand-held ultrasound devices and high-end scanners for real-time automatic measurements.
To distinguish between normal and pathological measurements of a given echocardiographic index, normal reference values based on healthy populations are needed. Echocardiographic measurements can vary between different ultrasound scanner models, imaging modalities, and workstation software. [7][8][9] Therefore, previously published reference values for manual measurements of mitral annular motion indices 10,11 may not be optimal for interpretation of specific automatic measurements.
Mitral annular motion is influenced by age and gender. [10][11][12][13] Indexing for body surface area has been suggested to be of less importance in recent studies. 11,14 Most studies have published categorical reference values for men and women in age intervals of at least 10 years. Continuous reference values based on age could lead to more precise diagnostics.
Thus, we aimed to create age-and gender-specific reference values for automatic measurements of MAPSE, S′, e′, and a′ ready for easy implementation in the everyday clinic. We also aimed to validate the automatic measurements by comparing them to manual measurements.

| Study population
The third wave of the Nord-Trøndelag Health study (HUNT3) 15

| Image acquisition
All participants underwent a complete transthoracic echocardiographic examination that was conducted according to contemporary guidelines 16,17 by one physician echocardiographer experienced in echocardiography. The acquisition protocol has been described previously. 10 Shortly, participants were examined in the left-lateral decubitus position with a Vivid 7 scanner (version BT06) using a phased-array transducer (M3S with bandwidth 1.5-4.0 MHz or M4S with bandwidth 1.5-3.6 MHz) (GE Vingmed Ultrasound). From the apical 4-chamber projection, separate B-mode (grayscale) and cTDI recordings optimized for evaluation of the left ventricle were obtained. The grayscale recordings had a mean frame rate of 44 per second. The Doppler frames in the cTDI recordings had a mean frame rate of 100 per second, and the underlying grayscale images had a mean frame rate of 25 per second. Echocardiographic data were stored digitally and subsequently analyzed.

| Manual measurements of mitral annular motion indices
The recordings were analyzed by two experienced physician echocardiographers. Echocardiograms were analyzed for M-mode-derived MAPSE ( Figure 1A) and cTDI-derived S′, e′, and a′ ( Figure 1B The temporal smoothing of the velocity traces was set to 30 ms.
The peak systolic velocity was set as the highest velocity observed during systole. The peak diastolic velocities were the minimum negative values during early and late diastole, both converted to absolute values.

| Automatic measurements of mitral annular motion indices
The apical 4-chamber cTDI recordings were transferred to a laptop computer, converted from DICOM (Digital Information and Communication in Medicine) to in-house file format and analyzed using the algorithm ( Figure 1C). The mean computation time for analysis of each recording was 1.1 seconds. The method has been comprehensively described previously. 1 Briefly, the measurements were acquired in three main steps: (a) Segmentation of the left ventricle on 2D grayscale frames, (b) cTDI-based tracking of the mitral annulus, and (c) analysis of cTDI-derived displacement and velocity curves. All automatic measurements were obtained from the septal and lateral side of the mitral annulus from three consecutive cardiac cycles and then averaged. All recordings were inspected to assess the mitral annular tracking. The tracking was judged as correct if the tracking points were located in the mitral annulus or basal myocardium near the insertions of the mitral valve leaflets for three consecutive heart cycles.

| Statistical analysis
Continuous variables are presented as mean ± standard deviation The study participants were stratified by gender and age (<40, 40-59 and ≥60 years). In each age group, the frequency of women and men in each age group was compared with the binomial test, and characteristics were compared with the independent t-test. The associations between (a) gender and age groups and (b) automatic measurements of mitral annular motion indices were tested by two-way analysis of variance, as all included cases had a complete set of measurements.
Homogeneity of variances was evaluated by dividing the maximal variance by the minimal variance for each mitral annular motion index, where a ratio < 3 was considered to indicate homogeneity. 18 For each index, homogeneity of variance was seen in all subgroups. P-values for differences in means between subgroups were Bonferroni-corrected.
Reference values based on gender and age groups were calculated as mean ± 1.96 SD, under the assumption that 95% of observations of a normally distributed variable can be expected to lie within such a range (prediction interval). The reference values for the gender-specific age groups are referred to as "categorical". The F I G U R E 1 Measurement acquisition. A, Manual MAPSE measurements from the septal (yellow line) and lateral (orange line) side by reconstructed M-mode. B, Manual measurements of S′, e′, and a′ from the septal (yellow curve) and lateral (orange curve) side by color tissue Doppler imaging. C, Graphical user interface for automatic measurements of MAPSE, S′, e′, and a′ by color tissue Doppler imaging. Only a grayscale frame is shown. a′ = peak late diastolic mitral annular velocity; e′ = peak early diastolic mitral annular velocity; LA = left atrium; LV = left ventricle; MAPSE = mitral annular plane systolic excursion; RA = right atrium; RV = right ventricle; S′ = peak systolic mitral annular velocity association between age and the mitral annular motion indices was further assessed by regression analyses. Treating age as a continuous variable in linear regression, reference values were created by estimating means with 95% prediction intervals, under the assumption that 95% of individuals of a certain age will have mitral annular motion measurements within a corresponding prediction interval.
These reference values are referred to as "continuous".
The automatic and manual measurements were compared in Bland-Altman analyses and by calculating the coefficient of variation (CoV). The CoV was calculated as the within standard deviation of the difference between the automatic and manual measurements divided by the corresponding mean. Means of automatic and manual measurements were compared with a paired t-test. The 95% limits of agreement were calculated as mean difference ± 1.96 SD. All the statistical analyses were performed with SPSS version 25.0 (IBM Corp).
P-values < .05 were considered statistically significant.

| Study population
Among the 1266 cases available for inclusion, 5 (0.4%) were missing cTDI recordings, 6 (0.5%) had too much image noise for manual measurement acquisition, and 1 (0.1%) was excluded due to rocking myocardial motion. In the 1254 remaining cases, the algorithm could not read the cTDI recordings in 5 (0.4%), and the algorithm tracked the mitral annulus incorrectly in 92 (7.3%). Thus, 1157 cases of which 608 (52.5%) were women were included in the final analyses (

| Relationship between gender, age and body surface area and the mitral annular motion indices
For the population as a whole, there was no significant difference in MAPSE between genders, but an interaction effect was found between gender and age groups (P < .001), where women had the largest differences in MAPSE between age groups. In both women and men, MAPSE was negatively correlated with age (r = −0.49. and −0.30, respectively, both P < .001). For the velocity indices, the relationship with age was similar for both genders. Women had lower S′ than men (6.6 ± 1.1 vs 6.9 ± 1.3 cm/s, P < .001). S′ was negatively correlated with age (r = −0.32, P < .001). Women had higher e′ than men (9.2 ± 2.5 vs 8.4 ± 2.5 cm/s, P = .009).
There was a strong negative correlation between e′ and age (r = −0.74, P < .001). Women had lower a′ than men (7.6 ± 1.5 vs 8.0 ± 1.6 cm/s, P = .001), and a′ was positively correlated with age (r = 0.38, P < .001). For the study population as whole, body surface area was uncorrelated with MAPSE (r = 0.06, P = .054), slightly positively correlated with S′ (r = 0.11, P < .001) and a′ (r = 0.14, P < .001), and slightly negatively correlated with e′ (r = −0.10, P < .001). Table 2 shows categorical reference values for automatic measurements according to gender and age groups. The categorical and F I G U R E 2 Distributions of measurements. Histograms depicting distributions of automatic and manual measurements of mitral annular motion indices in 1157 cases. Normal distribution curves are fitted to the histograms. Abbreviations as in Figure 1 continuous reference values are interleaved in Figure 3 for comparison. The largest discrepancies between categorical and continuous reference values were seen for e′.

| Agreement between automatic and manual measurements
The automatic measurements of MAPSE were lower than the manual measurements (P < .001), with mean difference ± SD −1.4 ± 2.0 mm.
There was no significant mean estimation bias between automatic and manual measurements of S′ (P = .13), with mean difference ± SD 0.0 ± 0.5 cm/s. The automatic measurements of e′ and a′ were slightly higher than the manual measurements (both P < .001), with mean difference ± SD 0.4 ± 0.7 cm/s and 0.6 ± 0.8 cm/s, respectively. Limits of agreement are shown in the Bland-Altman plots in Figure 4. The CoV was 11.7% for MAPSE, 5.5% for S′, 6.3% for e′ and 9.9% for a′. As indicated in Figure 4, the measurement differences  For S′, e′, and a′, both the automatic and the manual measurements were cTDI-derived. Indices derived by pulsed wave tissue Doppler imaging (pwTDI) and cTDI differ, with higher absolute values by pwTDI. 10,[21][22][23] In the European initiative Normal Reference Ranges for Echocardiography Study, only pwTDI indices of mitral annular velocities were presented. 24 Even though pwTDI spectra can be quickly interpreted, the possibility to place multiple regions of interest simultaneously is an advantage of cTDI. Furthermore, cTDI indices are less influenced by gain settings 22 and form an ideal platform for automatic evaluation of both systolic and diastolic LV function. For evaluation of diastolic function, pwTDI is often preferred, Note: Data are presented as mean ± standard deviation except for number of cases. Abbreviations: BMI = body mass index; BP = blood pressure; BSA = body surface area; n = number of cases; y = years.

| D ISCUSS I ON
but as cTDI is sensitive to subclinical changes in LV function 10,25,26 and carries prognostic information, 27  Abbreviations: a′ = peak late diastolic mitral annular velocity; e′ = peak early diastolic mitral annular velocity; MAPSE = mitral annular plane systolic excursion; S′ = peak systolic mitral annular velocity; SD = standard deviation. Other abbreviations as in Table 1. Abbreviations as in Figure 1 interpretation and higher diagnostic accuracy. However, there is  critical value of a t distribution with n-2 degrees of freedom (two-tailed at 95% confidence level). Other abbreviations as in Table 2.
F I G U R E 4 Agreement between automatic and manual measurements of mitral annular motion indices. Each dot represents one of the 1157 cases. The blue lines indicate the mean differences, and the red dashed lines indicate the 95% limits of agreement. Differences are computed as automatic minus manual measurements. Abbreviations as in Figure  1  S1). This indicates that the automatic and manual measurements demonstrated similar agreement across grades of normal systolic LV function and morphology.

| Study limitations
Most of the study participants were Caucasians of Northern European descent. Ethnicity may affect mitral annular motion indices. It is shown that a Chinese population had slightly different values compared with the Norwegian HUNT3 population. 13 Like in all population studies, selection bias could have occurred. All the examinations were conducted by one experienced echocardiographer using equipment by a single vendor, and two experienced observers analyzed the echocardiograms. This assured high quality of the recordings and analyses, but the agreement may have been better than if more observers with different experience and more vendors were included. The method for automatic measurements has not been released as a commercial software, and implementing the method on ultrasound scanners would require research and development.

| CON CLUS IONS
Reference values for automatic and manual measurements of mitral annular motion indices from a large, healthy population are presented. The automatic measurements showed good agreement with reference measurements by experienced echocardiographers.
Implementing automatic measurements of mitral annular motion and comparison with normal ranges in ultrasound scanners can allow for quick and accurate interpretation of LV function.

ACK N OWLED G M ENTS
This study was primarily funded by the Norwegian University of