Introduction: The stroke volume (SV) can be assessed by several echocardiographic methods. The type of SV and the diagnostic approach which will be used depends on the clinical scenario, especially in patients with valvular diseases. In case of normal cardiac geometry and the absence of valvular diseases the effective stroke volume corresponds to the forward stroke volume of the left (LV) and right ventricle (RV).

We hypothesized that in athletes with normal cardiac geometry and excellent acoustic windows left ventricular SV values obtained by Doppler echocardiography (SV_LVOT) will correspond to SV values assessed by other methods using 2D/3D echocardiography. The aim of the present study was to develop a deep learning algorithm for a plausible SV assessment.

Patients and methods: The study was performed in 46 competitive athletes (mean age 22 ± 4 years). The SV was measured in all athletes by eleven different methods:

effective left- and right ventricular SV (SV_LVOT/SV_RVOT) by Doppler echocardiography, SV_Teichholz by LV-M-Mode analysis, SV_monoplan using apical long axis, SV_4Ch, SV_2Ch and SV_biplan by biplane Simpson method using 4- and 2-chamber view, SV_triplan by triplane analysis and SV_{4D LVQ (auto)}, SV_{4D LVQ (man.)} and SV_{4D RVQ} by 3D volumetry. Mean values ± standard deviation and variances of SV are shown in Tab. 1. Statistical significance was accepted for p-value < 0.05.

Results: The highest stroke volume was found for SV_LVOT, the lowest for SV_{4D RVQ}, whereas SV_{4D LVQ (auto)} showed highest standard deviation (SD) and SV_{4D RVQ} lowest SD. The comparison of all methods showed no significant differences using univariate-analyses by ANOVA (p-value: 0.427).

Comparing two methods by t-test statistically significant differences were found between SV_LVOT and the following methods: SV_RVOT (p-value: 0.043), SV_monoplan (0.029) and SV_{4D RVQ} (0.004). No significant differences were found between SV_LVOT and the following methods: SV_Teichholz (p-value: 0.063), SV_4Ch (0.083), SV_2Ch (0.519), SV_biplan (0.315), SV_triplan (0.159), SV_{4D LVQ (auto)} (0.170) and SV_{4D LVQ (man.)} (0.086).

The mean SV value of all methods was determined in each athlete to estimate the accuracy of effective SV assessment resulting in SV_biplan closest to the mean value in 45 %, followed by SV_Teichholz in 31 % and SV_LVOT in 24 %.

Conclusions: In echocardiographic data sets with high image quality no significant differences of SV assessment were observed, whereas highest values were found for SV_LVOT. The assessment of SV values by LV or RV planimetry can be limited due to foreshortening views. Thus, SV_LVOT might probably be more robust and can be seen as the most appropriate parameter in future deep learning algorithms for verifiable SV quantification.
 

Table 1: Mean values ± standard deviation (SD) and variances of the different stroke volume (SV) measurements by 2D and 3D echocardiography in athletes.