Reference intervals for various measurements of canine left atrial size and function obtained using two-dimensional and three-dimensional echocardiography

Background: Many canine cardiac diseases are associated with left atrial (LA) remodeling and decreased function. For accurate assessment of LA indices, large-scale and prospectively determined reference intervals are necessary. Objectives: To generate reference intervals of LA size and function using two-dimensional and three-dimensional echocardiography. Animals: Two hundred and one healthy adult dogs. Methods: Left atrial volume was assessed in right parasternal long-axis, left apical four-chamber and two-chamber views using monoplane Simpson’s method, two-dimensional and three-dimensional speckle tracking. Additionally, LA diameter was measured in right parasternal short-axis and long-axis views. Furthermore, LA function was determined by measuring strain and calculating LA fractional shortening and ejection fraction. All variables were tested for correlation to heart rate, age, and body weight. For LA diameter and volume, scaling exponents and prediction intervals were generated using allometric scaling. Reference intervals for LA function parameters were calculated using nonparametric methods. Results: Left atrial diameter and volume showed a strong correlation with body weight. The scaling exponent for LA diameter was approximately 1/3 (0.34 e 0.40) and approximately one for volume measurements (0.97 e 1.26). Parameters of LA function showed no clinically relevant correlation with body weight, except for


Introduction
Many cardiac diseases in dogs are accompanied by changes in the left atrium (LA) during disease progression.Due to LA remodeling, an increase in LA volume and decrease in LA function can be observed [1e4].Recently, many studies have focused on changes in LA size and function and its importance in therapy and prognosis [2,3,5e9].
In dogs with preclinical myxomatous mitral valve disease, LA enlargement serves as one of the major indicators for initiation of therapy [10,11].As disease progresses, some LA measurements may be used to get a suspicion whether congestive heart failure is present or not [2,3,12].In addition, measurements of LA size and function are used as prognostic factors [7e9].Therefore, the assessment of the LA is an essential part of any echocardiographic examination.
Left atrial diameter (LAD) indexed to aortic diameter is currently the standard measurement of LA size in veterinary medicine [11,13].Due to the three-dimensional structure of the LA, enlargement cannot necessarily always be detected through linear measurement.Volumetric measurements of the LA have been recommended in human medicine for several years [14].Recently, three-dimensional assessment of the LA by echocardiography has also become possible.Some studies show a good agreement between three-dimensional transthoracic echocardiography and magnetic resonance imaging (MRI), which is considered to be the gold standard for assessment of cardiac dimensions [15,16].However, due to lack of reference values for three-dimensional echocardiography, it is not yet possible to use it in clinical practice.
Three-dimensional echocardiography can not only be used for measurement of LA volume (LAV), but also for assessment of left atrial function.Left atrial function, namely LA fractional shortening (LA-FS) and LA ejection fraction (LA-EF) can be calculated using LAD or LAV measurements.Alternatively, LA function can be measured directly by LA strain.A relatively new method to determine strain is speckle tracking [17].The advantage of this method over older methods of strain determination is its angular independence [18].In this regard, three-dimensional echocardiography provides the ability to track tissue Simpson's method of discs reflections (so-called "speckles") throughout the complete cardiac cycle even if their movement leaves a two-dimensional image plane [19].Therefore, the primary objective of our study was to establish reference intervals (RIs) for LAV and indices of LA function (strain, LA-EF, and LA fractional area change (LA-FAC)) by three-dimensional echocardiography.Because the specialized equipment required for this purpose is not available in daily clinical practice, alternative methods should also be considered.Accordingly, the subsidiary objectives were the following: (1) to establish RIs by calculating prediction intervals (PIs) for linear measurements as well as two-dimensional measurements by monoplane Simpson's method of disc (SMOD) and speckle tracking of the LA; and (2) to determine the repeatability of all measurements.

Study population
In this prospective study, dogs were included between February 2018 and September 2019.All dogs were privately owned dogs presented for a breeding examination, routine cardiological assessment, or as part of an offer for a free cardiac examination.The dogs had to be at least 12-monthold and had to have a body condition score between 4 and six out of nine.Dogs were regarded to be healthy based on the absence of owner-reported clinical signs, unremarkable complete physical examination, conventional echocardiography, and simultaneous electrocardiogram.The anatomy of the valves and the right heart had to be normal.Additionally, the left ventricular size was assessed in accordance to previously published RIs [20,21] and a LA to aortic root ratio 1.6 indicated that the LA was not enlarged.The maximal flow velocity of the aorta and pulmonary artery had to be < 2.2 m/s.Exclusion criteria were the following: (1) history of cardiac or respiratory disease; (2) heart murmur; (3) arrhythmia other than respiratory sinus arrhythmia; and (4) medications known to affect the cardiovascular system; (5) acquired or congenital heart diseases potentially leading to cardiac enlargement.Trivial insufficiency of the atrioventricular and semilunar valves was not an exclusion criterion if the auscultation was unremarkable and no morphological abnormality of the valve could be found.A heart murmur caused by an accelerated aortic flow velocity below 2.2 m/s was deemed acceptable if no other cause for the heart murmur could be found and the location of the murmur could be explained by an accelerated aortic flow velocity.
The study was approved by the Ethical Committee at the Department of Veterinary Medicine, LMU Munich (permission number 162-05-03-2019).

Conventional echocardiography
Echocardiography was performed by a boardcertified cardiologist (G.W.) or a resident under his direct supervision using an Epiq 7 a, b, c and d ultrasonographic system with various phasedarray transducers (1e12 MHz) according to the size of the dog.For the examination, the dogs were gently restrained in right and left lateral recumbence without sedation.A complete echocardiographic examination was undertaken in each dog as previously described [22] and echocardiographic loops were stored as digital data for separate offline analysis using TomTec Arena b software.A single investigator (J.T.) trained by the cardiologist (G.W.) performed all LA measurements using the TomTec software.

Linear measurements
For the linear measurements, two-dimensional loops of the right parasternal short-axis view (RPSA) at the level of the heart base, of the right parasternal long-axis four-chamber view (RPLA) optimized for the LA and of the right parasternal long-axis view optimized for the left ventricular outflow tract were recorded.The diameters of the LA in the RPSA were measured at two different time points.First, the maximal diameter of the LA was obtained one frame after aortic valve closure, corresponding to one to two frames after the T-wave on electrocardiogram and the minimal diameter was obtained at the beginning of the QRS-complex before aortic valve opening.The left atrial diameter in the RPLA was measured in the middle of the LA, parallel to the mitral valve as described previously [23].The measurements for the maximal diameter (LAD lax-max ) were performed one frame before mitral valve opening and for the minimal diameter one frame after mitral valve closure.Additionally, the aortic diameter in the right parasternal left ventricular outflow view was measured (Ao lax ) between the visible aortic valve leaflets when the aortic valve was maximally opened [23].Afterward, the ratio of the LA to aorta in the long-axis (lax) was calculated with the two different LAD values (maximal and minimal) a Philips Health Care, Germany.b TomTec, Unterschleissheim, Germany.
both divided by the Ao lax (i.e.maximal and minimal LA/Ao lax ).Furthermore, for both views (RPSA, RPLA) the LA-FS was calculated as follows: (LAD e minimal LAD)/maximal LAD s Â 100 ¼ LA-FS%.

Two-dimensional volume measurements
The modified monoplane SMOD was used for calculation of LAV from the RPLA, the left apical fourchamber view (4C) and left apical two-chamber view (2C) (Fig. 1) [24,25].All images were optimized for the LA and the frame rate was increased by decreasing depth and narrowing the image sector in order to optimize image quality.The inner edge of the LA border was manually traced excluding the LA appendage and the pulmonary veins.Thereafter, the TomTec software automatically determined the LAV using SMOD.
The measurement was performed at the following three time points:

Two-dimensional speckle tracking
The same three image planes used for LA SMOD measurements were additionally analyzed offline with two-dimensional speckle tracking using Tom-Tec software (two-dimensional cardiac performance analysis 1, Fig. 1) for assessment of LA size and function.The measurement time points for the maximal volume and the minimal volume were selected as described above.The endocardial border of the LA was manually traced at minimal volume after mitral valve closure by the point and click approach excluding the LA appendage and the pulmonary veins.Also, the fossa ovalis was excluded from the analysis in case of echo drop out.The software then automatically tracked these points during the cardiac cycle.For the maximal volume automatic tracking could be corrected manually if necessary.Based on this, the software calculated the maximal and minimal volume, the endocardial global longitudinal strain (LA-GLS), the LA ejection fraction (LA-EF), and the LA-FAC.
Three-dimensional echocardiography A X5-1 and X7-2 matrix probe were used for threedimensional echocardiography and a simultaneous electrocardiogram was recorded.The 4C view was used for three-dimensional images acquisition of the LA and a loop of two consecutive cardiac cycles were recorded, of which the better one was used for analysis.To optimize the image quality and increase frame rate, image sector and image depth were kept as narrow as possible.While acquiring loops for the LA, efforts were made that at least 70% of the left ventricle was visible, since the TomTec software package must first analyze the left ventricle before a correct three-dimensional analysis of the LA (4D LV-analysis 3 (LA)) can be performed.
First, the cardiac cycle to be analyzed was selected.Subsequently, the three anatomic landmarks of the left ventricle, such as the left ventricular apex, the mitral valve plane, and the aortic valve, were determined (Fig. 2).Then the maximal volume, pvolume, and minimal volume time points were set as described above.Based on this information the software automatically tracked the endocardial border of the left ventricle which could be manually corrected if necessary.Following this, the second landmark of the LA, the roof of the LA, was fixed, since the mitral valve plane was already determined at the definition of the left ventricle.The border of the LA was then automatically tracked and could be adjusted in 4C, 2C, left apical three-chamber, or short-axis view if deemed necessary.Last, the software calculated the following parameters of the LA: the maximal volume, the p volume, the minimal volume, the global longitudinal strain (LA-GLS 3D ), the LA-EF (LA-EF 3D ), and the active LA-EF (LA-trueEF 3D ).The active LA-EF corresponds to the LA-EF during the active contraction and was calculated as follows: (LAV at beginning p-wave e minimal LAV)/LAV at beginning p-wave Â 100 ¼ LA-active EF %

Repeatability
The images acquired from 10 randomly selected dogs were analyzed one each by three different trained investigators to assess interobserver measurement variability.For the intraobserver variability, one investigator (J.T.) measured all variables twice, with at least one week in between those two sessions.The investigators were blinded to previous measurements and each other's results.

Statistical analysis
Statistical analysis was performed with the open source software R c and the commercially available software MedCalc.d  The data were visually inspected and the Tukey method [26] was used to identify outliers.All outliers were thoroughly inspected and only excluded if image quality was too poor to generate reliable measurements.
All variables were tested for dependence on heart rate, age, and BW using Spearman's rank correlation coefficient.A correlation coefficient r > 0.4 was defined as potentially clinically relevant and consequently reported.
Variables with strong correlation (r > 0.7) with BW were normalized to BW using the allometric equation: In this equation, Y represents the measurement of LA, a the proportionality constant, and b the scaling exponent.For linear regression analysis, logarithmic transformation of the data was performed (log (Y) ¼ log (a) þ b Â log (BW)).As described by Cornell et al. [27], the subsequent, approximative formula was used to estimate the 95% PIs: Where log(a) is the constant determined by regression analysis, t is the Student's t value for n e 2 degrees of freedom and the desired confidence level, and S x,y is the standard error of the estimation [27].To calculate RI for specific body weights, the determined scaling exponent and the predicted 2.5 and 97.5 percentiles were inserted into the allometric equation along with the corresponding body weight.
If only a mild correlation (r > 0.4 and r < 0.7) with BW was shown, the population was divided into three groups (BW < 15 kg, BW ! 15 and <30 and BW !30).For each group, the median and the 95% prediction interval (PI) were calculated.If correlation with BW was not clinically relevant (r < 0.4), weight-independent reference values with the median and 95% PI were determined.For the estimation of the 95% PIs of these variables, the reference interval guidelines of the American Society for Veterinary Clinical Pathology [28] and the Clinical and Laboratory Standards Institute guidelines C28-A3 [29] were followed.If more than 120 measurements were available, 95% PIs of these variables were calculated using a nonparametric approach [29].If the population was subdivided into three groups and the sample size was between 40 and 120, the RIs were calculated using the robust method and the 90% confidence intervals of the limits were identified by bootstrapping of the data [30].A paired t-test was used to examine for significant differences between the measurements of the same variable in different image planes or between different measurement techniques.On the assessment of clinical relevance, the effect size was calculated using Cohen's D (dz) as the following: dz ¼ jdifference of means = pooled standard deviationj The following classification was made: dz > 0.80 large effect, dz 0.50e0.80moderate effect, dz 0.20e0.50small effect, and dz < 0.20 trivial effect.Trivial effects were defined as clinically irrelevant and not reported.
For intraobserver and interobserver measurement variability, the coefficient of variation (CV) was determined for each included dog individually.Then the mean value of the CV of all 10 dogs was calculated.For this purpose, the following formulas were used for calculation: CV ð%Þ ¼ standard deviation of measurement= mean of measurement Â 100 The repeatability was divided into the following grades: CV < 5% excellent repeatability, CV 5e15% good repeatability, CV 15e25% moderate repeatability, and CV > 25% low repeatability.
For all analyses, a P < 0.05 was defined as statistically significant.

Study population
This study included 211 dogs, with 10 dogs being excluded for poor image quality.This left us with a total number of 201 dogs.The study population consisted of 16 Australian shepherds, 15 Doberman pinscher, 14 Labrador retrievers, 10 dachshunds, seven golden retrievers, five boxers, four Chihuahuas, three Magyar Vizslas, three Yorkshire terriers, three Rhodesian ridgebacks, three Irish wolfhounds, three Border collies, 55 mixed-breed dogs and the remaining 60 dogs belonged to 50 different breeds.One hundred and twenty-three dogs were female of which 83 (67%) were spayed and 78 dogs were male of which 30 (38%) were neutered.The population showed a median BW of 23.0 kg (range 1.7e64.5 kg) and a mean age of four years and six months (range 1.0e14.1 years).The heart rate differed from 60 to 180 beats per minute with a median of 96 beats per minute.In 55 dogs, a trivial insufficiency of the mitral valve was detectable with color Doppler, however without an audible heart murmur and in absence of any morphological valve abnormalities.Moreover, a trivial insufficiency of the tricuspid, pulmonary, or aortic valve could be found in 54 dogs.Four dogs showed an accelerated aortic flow velocity with a maximum of 2.1 m/s and an audible heart murmur.

Left atrial measurement
All measurements of LAD, LAV, and aorta diameter showed a strong positive correlation with BW (Fig. 3).For all linear measurements, the exponent b was approximately 1/3 and for the volume measurements, b was around one.The exponents and constants for indexing the echocardiographic measurements to BW are presented in Table 1.The mean and the 95% PIs of the various LA measurements for dogs in different weight groups are presented in Table 2 (linear measurements), Table 3 (two-dimensional volume measurements), and Table 4 (three-dimensional and speckle tracking).The three-dimensional volumes are significantly smaller compared with the volumes calculated from twodimensional speckle tracking echocardiography (P < 0.01; effect size are dz ¼ 0.24 up to dz ¼ 1.12).No significant or clinically relevant differences are observed when comparing volumes generated by three-dimensional echocardiography and SMOD in the 4C and 2C views (P < 0.01 up to P ¼ 0.26, dz ¼ 0.19), except for LAV S-2C at p-wave (P < 0.01, dz ¼ 0.27) and minimal LAV S-2C (P < 0.01, dz ¼ 0.25) which are slightly larger and show a small effect.Measurements in the RPLA are larger than those in the 4C (P < 0.01) and 2C (P < 0.01) view and threedimensional echocardiography (P < 0.01), with a standard deviation of measurement moderate to large effect (dz ¼ 0.58 up to dz ¼ 1.29).Using the same measurement technique for 4C and 2C volumes, no clinically relevant variations are detected (P ¼ 0.04 to P ¼ 0.36, dz ¼ 0.15).Comparing the different techniques, the volumes calculated with SMOD are smaller than the volumes measured by two-dimensional speckle tracking (P < 0.01, dz ¼ 0.26 up to dz ¼ 0.38), beside the RPLA view, where the SMOD volume is larger (P < 0.01, dz ¼ 0.39).A detailed overview of mean differences and the confidence intervals of the difference are listed in Table 5.
The variables LA-EF 3D and LA-GLS ST-2C showed a mild negative correlation with BW (Fig. 3).Therefore, the population was separated into three body weight classes and, for each class, the median and the 95% PI was calculated using the bootstrap approach.The results are summarized in Table 6.
For all other parameters of LA function (LA-GLS, LA-EF, LA-FS, and LA-FAC) and the ratios of LA to aorta, no significant or no clinically relevant correlation with BW (r < 0.4) could be detected.Consequently, weight-independent reference values were calculated.The statistic results with median and 95% PIs are shown in Table 7.
None of the LA or aortic measurements demonstrated a clinically relevant association with age or heart rate.

Repeatability
The linear measurements and the LA to aorta ratios showed excellent intraobserver and good interobserver repeatability.LA FS in the RPSA and RPLA demonstrated good intraobserver measurement repeatability, but only a moderate interobserver repeatability.Intraobserver repeatabilities for SMOD, three-dimensional and two-dimensional speckle tracking measurements were excellent to good excluding LA-trueEF 3D .Repeatabilities for interobserver measurements were also good to excellent with exception of LA-trueEF 3D , LAV ST-2C- max , and LAV ST-2C-min .
The coefficient of variation of the LA-trueEF 3D in 10 randomly selected dogs was 22.54% for intraobserver repeatability and 17.85% for interobserver repeatability.When the measurement of a single subject with the lowest agreement was excluded, the coefficient of variation improved to 10.89% for intraobserver repeatability and 9.35% for interobserver repeatability.The interobserver repeatability of maximal LAV ST-2C was 15.73% and 20.04% for minimal LAV ST-2C .
A detailed overview of the correlation coefficients is listed in Table 8.

Discussion
This study evaluated various linear, two-dimensional, and three-dimensional measurement methods for LA diameter, volume, and function in a large cohort of 201 healthy adult dogs.Reference intervals of numerous LA measurements not previously reported were generated including three-dimensional volume and function of the LA, LAV measured by monoplane SMOD in the 2C and 4C view, minimal LAD in the RPLA and LA volume and function measured by two-dimensional speckle tracking.In addition, some previously described RIs were reevaluated.For the strongly body weight-dependent parameters LAD and LAV, an exponent was determined by means of allometric scaling in order to normalize the values for daily use.As previously described, the exponent in our case was also close to 1/3 for linear measurements (0.34e0.40) and approximately one for volumetric measurements (0.97e1.26) [24,27,31].Most values of LA function showed no correlation with body weight, a mild negative dependence was observed in only two measurements (LA-EF 3D and LA-GLS ST-2C ).
In human medicine it has been shown that threedimensional echocardiography is the best echocardiographic alternative to MRI for LA assessment [15].Further studies also showed that three-dimensional LA volume is a better predictor for cardiovascular events in humans than volumetric measurements acquired via two-dimensional biplane methods [32,33].However, three-dimensional echocardiography also presents some difficulties.In 10 of our 211 dogs, the three-dimensional loops could not be evaluated.One reason for this was a frame rate that was too low for the image analysis, meaning a frame rate of at least one quarter of the heart rate (in beats per minute) is necessary.Other reasons were too large data volumes of the recorded loop, which resulted in a failure of analyzability, or a poor image quality, meaning that the LA endocardial borders could not be detected.For prevention of these limitations, only two cardiac cycles were recorded, and the image sector was selected as small as possible.This increased image quality resulted in a portion of the left ventricular apex not being included in the image.One should consider that for threedimensional evaluation of the LA and the left ventricle, two different image recordings are necessary, as the evaluation of both structures within one image is currently not feasible.With further development of three-dimensional ultrasound technology, this may no longer be necessary in the future.Reference values of cardiac dimensions are mandatory for the use of a measurement method in order to detect abnormalities.In our study, the three-dimensional volumes are slightly smaller   2C: left apical two-chamber; 3D: three-dimensional echocardiography; 4C: left apical four-chamber; LAV: left atrial volume; max: maximum; min: minimum; p: volume before atrial contraction; R: right parasternal long-axis; ST: two-dimensional speckle tracking.than volumes calculated from two-dimensional speckle tracking echocardiography and are comparable to the volumes generated with SMOD in the 4C and 2C view.This corresponds to the results of a study by LeBlanc et al. [34] where no differences between monoplane SMOD or biplane area-length method in the apical views and threedimensional volume of the LA was detected.However, compared to the biplane area-length reference values from a previous study, our three-dimensional volumes are slightly larger [31].This is in contrast to a newer study which    compared monoplane SMOD, three-dimensional, and MRI and demonstrated that monoplane SMOD usually overestimates the volume measured by MRI, whereas it is underestimated by threedimensional measurements, resulting in the twodimensional volume being the largest [16].These results are similar to another study performed in cats using the TomTec software to assess LA dimensions, which also showed smaller threedimensional SMOD acquired volumes than twodimensional SMOD acquired volumes [35].These different results may be due to the different technical equipment and measurement methods.Therefore, if possible, reference values that fit the existing equipment should be used.Three-dimensional echocardiography is not feasible in every clinical setting, as it requires special technical modalities and corresponding software.Therefore, we have additionally evaluated LAV measurements by two-dimensional methods.Left atrial volume was determined by monoplane SMOD and two-dimensional speckle tracking in the RPLA, 4C, and 2C views.When comparing the results between SMOD and speckle tracking of the corresponding ultrasound planes, the same difference could be found and the speckle tracking volumes were slightly larger than the volumes obtained by SMOD.Furthermore, when looking at the various planes amongst each other, the volumes are slightly different, with LAV being largest in the RPLA view and smaller in the 4C and 2C view.Previous studies showed similar results for LAV being larger in the RPLA than in the 4C view [24,34].Consequently, the volumes are not comparable and the corresponding reference values must be used for each image plane and cannot be used interchangeably.There is only one study to date that published RIs for monoplane volume measurements of the LA, in which LAV was determined in the RPLA in 122 dogs [24].The reported 97.5 percentile for LAV S-R-max of 1.62 mL/ kg is slightly smaller than our results (around 1.8 mL/kg).This small difference can probably be attributed to the different scaling exponents [24].We determined the RIs by allometric scaling and not just by normalization of body weight.It is easier to only normalize the measurement to body weight, but allometric scaling gives more accurate results especially when the exponent is further from one (e.g.1.26 for minimal LAV S-R ).Therefore, all variables with a strong correlation with body weight were reported with the corresponding exponent and a 95% PI in this study.
In human medicine, the use of the biplane methods for volume measurements is recommended when three-dimensional echocardiography is not available [14].Following this, Ho ¨llmer et al. [31] presented reference values for LAV measured by biplane area-length method.However, for biplane methods it is important to generate two different views, usually 2C and 4C view, with a good image quality and the correct image plane.This is not always possible in daily practice, so the determined monoplane reference values should allow assessment of LAV in each patient.A comparison of the results with Ho ¨llmer et al. [31] showed that our study determined larger volumes.For example, LA enlargement in a 10 kg dog in Ho ¨llmer et al. [31] is predicted at a maximal volume above 9.3 mL, whereas, in this study, a volume up to 14.9 mL in the 4C view is considered normal.This disagreement may be due to different measurement techniques and this assumption is supported by a study by LeBlanc et al. [34].In this study, LA measurement using monoplane SMOD generated larger volumes than the biplane area-length method.
Several studies already investigated the LAD and indices of LA to aorta in the RPLA and RPSA and determined reference values.Comparing the linear maximal and minimal measurements of the LAD in RPSA with a previous study [36], similar results are obtained.Various reference values can be found in the literature for long-axis maximal LA to aorta ratio from long-axis.Some studies [36e38] determined the upper reference limit to be 2.1, whereas another study [23] found a higher limit (2.4), similar to our results (2.5).This discrepancy may be attributable to different measurement methods, since the studies with lower values used a different aortic measurement method than that used in the present study.Our maximal LAD results also show a slight difference (exponent 0.35 and 97.5 percentile 1.38) when compared to previously published studies [24,36,39].The other studies calculated a lower exponent and a higher 97.5 percentile and the cause for this is unclear, but it may be related to minimal variations in the imaging planes.Moreover, to the best of the author's knowledge, this is the first study to evaluate LA/Ao lax-min and LAD lax-min and define RI for clinical use.
In addition to volume measurements, the assessment of LA function becomes increasingly important in human medicine [6,40].Therefore, one of our aims was to determine the normal function parameters in healthy dogs.In some studies, PIs for LA function parameters have already been established, but these showed some disadvantages.Most of these studies only had a small study population between 23 Reference intervals for various left atrial measurements and 80 dogs.One study examined 120 dogs of which 97 dogs weighed less than 10 kg, so the larger dogs were not adequately represented [41].Therefore, this is the first study using a large number of dogs (201) and a wide range of body weights (1.7e64.5 kg).For LA-FS, the calculated RIs in the RPLA and RPSA were very similar and comparable with previously described reference intervals [38].The LA-EF 3D and LA-GLS ST-2C showed a mildly negative correlation with body weight (r ¼ À0.451/-0.449).Consequently, prediction intervals for all dogs and additionally, PIs for three body weight classes were calculated.The weight classes were arbitrarily set so that each group contained at least 40 dogs, as this is recommended for calculating RIs [28].All remaining function parameters showed no clinically relevant correlation with body weight.Comparing the results for the three image planes (RPLA, 4C and 2C) for LA-EF, LA-GLS, or LA-FAC, significant differences with small to moderate effect could be found.Therefore, the corresponding reference values for each image plane should be used to evaluate the parameters of LA function.However, all variables show wide distribution and a large range of normal values.Thus, further studies are needed to investigate their utility in clinical use.

Limitations
Despite the prospective study design, the study contains some limitations.Blood pressure measurement and a complete blood testing to exclude asymptomatic disease were not performed.However, the classification as a healthy dog without systemic disease was based on several factors (owner's statement, history, physical constitution, clinical examination, and echocardiography).
The American Society for Veterinary Clinical Pathology [28] recommended a minimum of 40 dogs to generate RIs.Due to the small number of dogs per breed the influence of the breed on measurements was not evaluated and no breed-specific RIs could be calculated.Additional studies are necessary to generate breed-specific RIs for the LA.Because we studied a large number of different dogs, it is still possible to use RIs provided by this study, but a breed-related variation should be kept in mind.
The three-dimensional measurement of the LA was only performed once.To determine new RIs, the measurement is usually repeated in three consecutive cardiac cycles and the averaged results of these three cycles is then used.The three-dimensional measurement itself is more time-consuming than previous measurement methods, therefore repetition is not practical and feasible in daily routine.This study showed a good repeatability for three-dimensional echocardiography even without repetition and is therefore routinely applicable.
All measurements were determined in one institution using the TomTec software.Especially for strain and three-dimensional measurements, it must be assumed that the software and the ultrasonographic system might influence the measurement results.
A further limitation is the fact that the obtained reference intervals could not be compared to MRI as a gold standard for volume assessment.Given the fact that the study was performed on 201 client-owned healthy dogs and the need for anesthesia, an MRI was not feasible in this study.
Finally, it should be noted that a different threshold in terms of correlation with BW may lead to a different calculation of the reference values.

Conclusions
This study established RIs for various left atrial parameters including those derived using threedimensional echocardiography.The wide range of described measurement methods offers each user the opportunity to select the appropriate reference values for the assessment of the LA depending on the available equipment and technical experience.Currently, linear parameters and SMOD volumes are particularly suitable for clinical practice because they are easy to perform and can be measured by most ultrasound equipment.Whereas speckle tracking and three-dimensional echocardiography are reserved for specialists for the time being, it may have the potential to be used more widely in the future.Further studies are needed to evaluate the utility of these parameters in dogs with heart disease and to determine the best measurement method for the identification of LA changes.

Figure 1
Figure 1 Measurement of the left atrial volume in the left apical four-chamber view using monoplane Simpson method of discs (A) and two-dimensional speckle tracking (B and C); definition of the left atrial endocardial border at the minimal volume for two-dimensional speckle tracking (B) and the results of the automatic tracking presented as a strain curve (C).EDV: enddiastolic volume; EF: ejection fraction; EndoGCS: endocardial circumferential strain; EndoGLS: endocardial global strain; ESV: endsystolic volume; FAC: fractional area change.

Figure 2
Figure 2 Three-dimensional echocardiography of the left atrium.Definition of anatomic landmarks of the left ventricle (A) and the left atrium (B).Automatic calculation of the left atrial volume presented as a threedimensional model (C) and a volumetric time curve (D).

Figure 3
Figure 3 Scatter diagram with a trend line for correlation between LAD sax-max in mL (A), maximum three-dimensional LAV in mL (B), LA-EF 3D in % (C), LA-EF S-R in % (D) and body weight in kg.3D: three-dimensional echocardiography; LAD: left atrial diameter; LA-E: left atrial ejection fraction; LAV: left atrial volume; max: maximum; R: right parasternal long-axis; S: Simpson's method of disc; sax: right parasternal short-axis.

Table 1
Results of regression analyses and constants for indexing left atrial measurements including prediction intervals.

Table 2
Mean values and 95% prediction intervals (in centimeters) of left atrial linear measurements in dogs.

Table 3
Mean values and 95% prediction intervals (in mL) for Simpson's method of disc left atrial volume measurements in dogs.

Table 4
Mean values and 95% prediction intervals (in mL) for 3D echocardiography and 2D speckle tracking of left atrial measurements in dogs.

Table 5
Differences of the measurement methods.

Table 7
Median and 95% prediction intervals for weight-independent left atrial measurements in dogs.

Table 6
Median and 95% prediction intervals [No. of obs.] of those left atrial measurements showing a mild correlation with body weight in dogs.

Table 8
Coefficient of variation for intraobserver and interobserver measurement variation of left atrial and aortic measurements.
a Measurement was not feasible in five out of 10 dogs.b Values displayed in the parentheses resulted when the measurement with the lowest agreement was excluded.