This study showed that the automated 3DE measurement, assisted with few manually adjusted Heartmodel images (13/97), was highly accurate in measuring LV volume and EF in patients with a variety of heart diseases. In addition, it was more timesaving and reproducible than the manual 3DE. The LVEDV and LVESV values obtained by automated 3DE were higher compared to manual 3DE, and there was a good consistency in assessing the impairment degree of systolic function in patients with wall motion abnormalities with these two technologies.
The Heartmodel software was designed to automatically measure LV volume and EF in order to reduce subjective factors, increase reproducibility and save time. However, previous studies on automated 3DE demonstrated a large amount of manual editing following after the automated contouring [5,6,7,8,9,10,11]. The purpose of fully automated measurement was therefore not clearly achieved, and the advantages of the new technology were not presented totally. Our study tended to realize the automated measurement to the greatest extent by rationally setting the default endocardial border values. Among the 97 enrolled patients, only 13 images required manual editing. Special attention should be paid to the fact that all the 8 patients with apical wall motion abnormalities were incorrectly automated contoured (Fig. 2). The reason may be that the apical myocardium located in the near field of ultrasound was not clear in apical 4-chamber view because of reverberation artifact and focus location which was set at the mitral valve-papillary muscle level. This in turn affected the recognition and tracking of the endocardium in the apex region by Heartmodel software. We suppose that the wall motion of apical segments was approximately estimated on the basis of basal and midventricular segment wall motion. Our conjecture was supported by the following phenomena: for patients with global wall motion abnormalities or without wall motion abnormalities, the software can correctly outline the apical endocardium because the amplitude of LV wall motion in these cases was essentially the same in all segments, whilst for patients with regional wall motion abnormalities in the apical segments, the Heartmodel software seemed to outline the apical endocardium according to the basal and midventricular segment wall motion instead of tracing the true position of apical endocardium. Therefore, when apical wall motion was inconsistent with the basal and midventricular segment motion, attention should be paid whether manual adjustment was needed in apical segments. The other 5 patients who needed manual adjustment consisted of 3 patients with hypertrophic cardiomyopathy and 2 patients with prominent LV wall thickening due to hypertensive heart disease or aortic valve stenosis. Uneven wall thickening and hypoechoic myocardium close to the endocardium accounted for the inaccurate automated contouring. According to our study, fully automated quantification of LV volumes and systolic function using Heartmodel software could be achieved in most patients and manual adjustment was needed merely in the minority of the patients.
In this study, the default endocardial border values were set as 66% and 40% for ED and ES, respectively. EDV and ESV obtained by automated 3DE were higher compared to manual 3DE, and the bias for EDV and ESV were 16 ± 18 ml and 11 ± 12 ml, respectively. The different methods of determining the endocardial border with the automated and manual 3DE accounted for the biases: On automated 3DE, although the endocardial border cannot be clearly explored in most patients, the more robustly recognized inner and outer extents of the myocardial tissue, i.e., the interface of the blood-tissue and the compacted myocardium can be easily detected by Heartmodel software. In our study we obtained the default endocardial border values from high quality heartmodel images, which represented the relative position of the endocardial border between the inner and outer extents of the myocardial tissue with varying heart shapes and sizes. Then all the subsequent Heartmodel images were automated analyzed with the determined default border values. So technically we could count trabeculae muscle in the LV volume on automated 3DE regardless of the image quality. On the contrary, on manual 3DE, although the trabeculae was planned to be counted in LV volume, the spatial resolution in most patients was insufficient to clearly define the endocardial trabeculae, which was, as a result, lumped together with the myocardium rather than being included in the LV cavity. This was the most significant potential source of volume underestimation by 3DE [14]. Most previous studies but two [6, 11] supported that the LV volumes obtained by automated 3DE were higher than those by manual 3DE [5, 8,9,10]. In addition, automated 3DE and CMR were compared in three studies, and results showed that the LV volumes were still underestimated by automated 3DE [7, 8, 11]. A meta-analysis of 34 studies reported that the overall pooled biases of manual 3DE were −19.1 ± 17.1 ml and −10.1 ± 14.9 ml for EDV and ESV compared with CMR [3]. According to our results, the measured values using automated 3DE might be very close to that of CMR values. A multicentric study also came up with the result that the automated 3DE measurements were closer to that of CMR than manual 3DE [15].
Practically, EF is the most important index in assessing the cardiac function for clinicians. Previous studies have reported that automated 3DE and manual 3DE were highly consistent in measuring EF. While some studies suggested no significant difference between the two methods [5, 6, 11], two studies found that the EF measured by automated 3DE was slightly lower than that measured by manual 3DE [8, 10]. Our study arrived at similar result with the bias −1%, which might be explained by the fact that the frame rate of full volume image was slightly higher than that of Heartmodel images. The impairment degree of EF in patients with wall motion abnormalities was further classified in the present study and the results showed that the automated 3DE with manual adjustment showed better correlation with manual 3DE than that without adjustment. Without manual adjustment, automated 3DE tended to overestimate the EF measurement in patients with regional kinetic abnormalities of the apex and underestimate it in patients with prominent LV wall thickening.
Manual 3DE has been considered the most reproducible ultrasonic technology for measuring LV volume and EF so far [16]. According to our results, the reproducibility among inter- and intra-observer using automated 3DE was better than those using manual 3DE in measuring EDV, ESV and EF. This was consistent with a multicentric validation study describing both inter- and intra-observer variability were lower for the automated measurements than conventional manual technology for all parameters [6]. The lower variability of the automated method might be explained by the fact that the automated 3DE measurement did not require much experience to operate and automated analytic method was more objective.
This study has some limitations. Firstly, CMR was currently recognized as the gold standard for measuring LV volume and EF; however, manual 3DE rather than CMR was used as a reference standard in our study. Although the volume tended to be underestimated, the accuracy of manual 3DE was proved to be comparable with that of CMR [3]. In order to warrant the accuracy as standard, all of full-volume images for manual 3DE were analyzed by an experienced physician. Secondly, our sample size was not large enough to include all kinds of heart diseases, and further research is still needed to improve the results.