以心脏磁共振（CMR）为金标准，探讨基于机器学习的三维超声心动图（ML-3DE）全自动测量左心室容积和射血分数的准确性及影响因素分析。

连续前瞻性地纳入2021年1月至2月于华中科技大学同济医学院附属同济医院行CMR的患者58例，在CMR检查后的24 h内采集三维超声心动图图像，采用ML-3DE全自动测量左心室容积与功能参数，包括左心室舒张末期容积（LVEDV）、左心室舒张末期容积指数（LVEDVi）、左心室收缩末期容积（LVESV）、左心室收缩末期容积指数（LVESVi）、每搏输出量（SV）及左心室射血分数（LVEF）。对ML-3DE的测值与CMR的测值进行相关性和一致性分析，单因素分析及逐步多元线性回归分析探讨左心室结构与功能的异常特征（非对称性左心室肥厚、LVEF减低、左心室扩大、节段性室壁运动异常）对ML-3DE与CMR测值差异（ΔLVEDV、ΔLVESV）的影响。

53例（91%）患者完成了ML-3DE对左心室容积与功能的定量分析。ML-3DE测量的左心室各参数指标组内相关系数（ICC）值均＞0.85。左心室各参数测值的ML-3DE与CMR测量值均具有中等以上相关性（LVEDV、LVEDVi、LVESV、LVESVi的相关性系数r=0.94、0.91、0.95、0.93；而LVEF、SV的相关性系数r=0.65、0.61；P均＜0.01）。Bland-Altman法表明2种方法测值一致性较高（90%数据位于Bland-Altman95%一致性区间内），但ML-3DE测量的左心室容积参数（LVEDV和LVESV）呈不同程度的低估，与CMR测值比较，差异有统计学意义（P均＜0.05）。逐步多元线性回归分析发现，LVEF减低、左心室扩大、非对称性左心室肥厚是ML-3DE与CMR测量LVEDV差异的独立影响因素；非对称性左心室肥厚、节段性室壁运动异常是ML-3DE与CMR测量LVESV差异的独立影响因素。

ML-3DE实现了左心室容积与功能参数的全自动定量分析，与CMR金标准具有较好的相关性和一致性；LVEF减低、左心室扩大、非对称性左心室肥厚、节段性室壁运动异常是ML-3DE与CMR左心室容积测量差异的独立影响因素。

To investigate the accuracy and influencing factors of fully automatic measurement of left ventricular volume and ejection fraction by machine learning-based three-dimensional echocardiography (ML-3DE) with cardiac magnetic resonance (CMR) as the gold standard.

Fifty-eight patients who were underwent CMR at Tongji Hospital Affiliated to Tongji Medical College of Huazhong University of Science and Technology from January 2021 to February 2021 were prospectively enrolled. Three-dimensional echocardiography was performed within 24 hours after CMR in the 58 patients. Then, ML-3DE was used to automatically measure left ventricular volume and functional parameters, including left ventricular end-diastolic volume (LVEDV), left ventricular end-diastolic volume index (LVEDVi), left ventricular end-systolic volume (LVESV), left ventricular end-systolic volume index (LVESVi), stroke volume (SV), and left ventricular ejection fraction (LVEF). Pearson correlation and Bland-Altman analyses were performed to compare the LV measurements derived from AI-based 3DE and CMR. Univariate analysis and stepwise multiple linear regression analysis were used to investigate the effects of abnormal characteristics of left ventricular structure and function (asymmetric left ventricular hypertrophy, reduced LVEF, left ventricular enlargement, and segmental wall motion abnormalities) on the differences between ML-3DE and CMR (ΔLVEDV and ΔLVESV).

Fully automatic measurement of left ventricular volume and function by ML-3DE was performed in 53 (91%) patients. The interclass correlation coefficient (ICC) of left ventricular parameters measured by ML-3DE was >0.85. The LV parameters measured by the two methods had moderate correlations; the correlation coefficients of LVEDV, LVEDVi, LVESV, and LVESVi were 0.94, 0.91, 0.95, and 0.93, while the correlation coefficients of LVEF and SV were 0.65 and 0.61, respectively (P＜0.01). Although Bland-Altman analysis showed a good consistency between the two methods (90% of the data were located in the 95% limits of agreement), the left ventricular volumes (LVEDV and LVESV) measured by ML-3DE were underestimated, which were statistically different from those measured by CMR (P ＜0.05). Stepwise multiple linear regression analysis showed that the reduced ejection fraction, dilated left ventricle, and asymmetric left ventricular hypertrophy were independent factors affecting the difference of LVEDV measured by CMR and ML-3DE, and the asymmetric left ventricular hypertrophy and segmental wall motion abnormality were independent impact factors affecting the difference of LVEDV measured by CMR and ML-3DE.

ML-3DE allows for fully automatic quantitative analysis of left ventricular volume and function, showing a close correlation and consistency with CMR (gold standard). Reduced LVEF, left ventricular enlargement, asymmetric left ventricular hypertrophy, and segmental wall motion abnormality are independent factors affecting the difference in LV volume measurements between ML-3DE and CMR.