Artificial Intelligence in Clinical Echocardiography: Many Expectations, but Deep Uncertainties for Defining Strategies to Overcome Difficulties and Obstacles

From ejection fraction, to myocardial strain, and myocardial work in echocardiography: Clinical impact and controversies

“…If available, 3‐dimensional (3D)‐based LV EF evaluation may be accurate and reproducible, in the presence of good ultrasound window, taking into account the lower temporal resolution as well 3 . Moreover, artificial intelligence (AI)‐assisted echocardiography may represent another tool in order to improve the reproducibility 5–7 …”

“…In the real life, effectively, the need to evaluate the individual subject in the emergency setting, 5 taking into account the worst possible clinical and logistic conditions (e.g., in the recent COVID‐19 pandemic setting, 42,43 in “normal” routine clinical practice in the emergency department, as well as in outpatient clinical management 5,6 ), has to be considered. Echocardiographic LV EF provides a first‐line (generally and historically considered) sufficiently reliable and reproducible LV global systolic function estimation in the acute setting, in order to provide an immediate “eyeshot” of the clinical picture, along with the possibility of fast reassessment after medical treatment initiation, titration, and optimization 3,4,8–10 .…”

“…10,13,25,26 Combining strain imaging with ultrasensitive troponin has been proposed to overcome that limitation. 13,25,26 Finally, other GLS limitations include the important image-quality relation, 3,5,6,14 intervendor/intersoftware variability, 3,5,6,14 age and load dependency, 3 and chest shape 24 (Table 1).…”

“…2,32,36 Those limits are already present in nonacute cardiac assessment, more and more in complex and critical patients. 2,3,5,6,14,24,32,36 Moreover, MW is not completely independent of load, being derived by myocardial strain measurements as well. 2,39 Also, the assumption that a particular percentage deterioration in a given MW parameter is causing similar proportional clinical consequences as that same percentage worsening in another MW parameter has to be demonstrated yet.…”

“…Optimizing image quality and frame rate is crucial 3 . Averaged GLS measurements should be made in standard apical long‐axis 4C, three‐chamber (3C), and 2C views, 2,3 and generally calculated using midwall deformation, 3 also with AI‐enhanced software if available/feasible 5–7 . 3D ST could eliminate the problem of through‐plane myocardial motion, inherent characteristic of 2D imaging, taking into account that 3D strain is currently limited by low frame rates 3 .…”