Artificial Intelligence in Echocardiography: The Time is Now

Abstract

Artificial Intelligence (AI) has impacted every aspect of clinical medicine, and is predicted to revolutionise diagnosis, treatment and patient care. Through novel machine learning (ML) and deep learning (DL) techniques, AI has made significant grounds in cardiology and cardiac investigations, including echocardiography. Echocardiography is a ubiquitous tool that remains first-line for the evaluation of many cardiovascular diseases, with large data sets, objective parameters, widespread availability and an excellent safety profile, it represents the perfect candidate for AI advancement. As such, AI has firmly made its stamp on echocardiography, showing great promise in training, image acquisition, interpretation and analysis, diagnostics, prognostication and phenotype development. However, there remain significant barriers in real-world clinical application and uptake of AI derived algorithms in echocardiography, most importantly being the lack of clinical outcome studies. While AI has been shown to match or even best its human counterparts, an improvement in real world outcomes remains to be established. There are also legal and ethical concerns that hinder its progress. Large outcome focused trials and a collaborative multi-disciplinary effort will be necessary to push AI into the clinical workspace. Despite this, current and emerging trials suggest that these systems will undoubtedly transform echocardiography, improving clinical utility, efficiency and training.

The Role of Artificial Intelligence in Echocardiography

“…This requires manual tracing of the end-systolic and end-diastolic contours in the apical four- and two-chamber views [ 26 ]. These methods and techniques for tracing biplane disc summation are subject to significant variability and have poor correlation with the gold standard Cardiac MR (CMR) [ 3 , 27 ].…”

“…The model was able to automatically identify standard apical views, time cardiac events, and measure GLS across a variety of cardiac conditions. It demonstrated minimal differences between the two methods, with an absolute difference of 1.8% [ 3 , 35 ]. The method utilized was rapid, taking less than 15 s per study with AI compared to 5–10 min with the conventional method [ 35 ].…”

“…While artificial intelligence has been around since the 1950s, in recent years there has been a strong focus on the application of AI with respect to diagnostic imaging. Machine learning and other AI techniques can recognize a variety of patterns within the imaging modalities, particularly echocardiography [ 3 ]. Echocardiography involves moving frames and can be more challenging to train AI than still images, such as those from Computed tomography (CT) or Magnetic resonance imaging (MRI) sequences.…”

“…Combining clinician interpretation with information obtained from machine learning algorithms can refine the accuracy of echocardiography. By combining these two parameters, this will reduce inter- and intra-operator variability [ 3 ]. Moreover, this can provide additional predictive information that may be too obscure for the human eye to detect.…”

Emerging Role of Artificial Intelligence in Echocardiography

“…A concerted multidisciplinary effort with engineers, computer scientists, sonographers, and physicians will be necessary to introduce AI in clinical echocardiography. [ 12 ]…”