Automated quantification of mitral valve anatomy using anatomical intelligence in three-dimensional echocardiography

Quantitative analysis of mitral valve morphology with three-dimensional (3D) transesophageal echocardiography (TEE) provides anatomic information that can assist clinical decision-making. However, routine use of mitral valve quantification has been hindered by tedious workflow and high operator-dependence. The purpose of this paper was to evaluate the feasibility, accuracy and efficiency of a novel computer-learning algorithm using anatomical intelligence in ultrasound (AIUS) to automatically detect and quantitatively assess the mitral valve anatomy.

A novice operator used AIUS to quantitatively assess mitral valve anatomy on the 3D TEE images of 55 patients (33 with mitral valve prolapse, 11 with functional mitral regurgitation, and 11 normal valves). The results were compared to that of manual mitral valve quantification by an experienced 3D echocardiographer and, in the 24 patients who underwent mitral valve repair, the surgical findings. Time consumption and reproducibility of AIUS were compared to the manual method.

AIUS mitral valve quantification was feasible in 52 patients (95%). There were excellent agreements between AIUS and expert manual quantification for all mitral valve anatomic parameters (r=0.85-0.99, p<0.05). AIUS accurately classified surgically defined location of prolapse in 139 of 144 segments analyzed (97%). AIUS improved the intra- [intraclass-correlation coefficient (ICC)=0.91-0.99] and inter-observer (ICC=0.86-0.98) variability of novice users, surpassing the manual approach (intra-observer ICC=0.32-0.95; inter-observer ICC=0.45-0.93), yet requiring significantly less time (144±24s vs. 770±89s, p<0.0001).

Anatomic intelligence in 3D TEE image can provide accurate, reproducible, and rapid quantification of the mitral valve anatomy.