Artificial intelligence (AI) is quickly becoming a buzz word in medicine. Numerous companies and researchers are investing resources to develop and test AI technology and tools. But what is AI? And how is it different from machine learning or deep learning? How will AI impact the future of cardiac imaging?

AI, in broad terms, comprises systems that allow computers or machines to perform or mimic human thinking. As it relates to cardiac imaging, AI enables automated recognition and segmentation of cardiac structures and performs quantification of cardiac chambers using pre-specified rules, algorithms, or instructions. Furthermore, AI may improve patient selection, image acquisition and reconstruction, and identification of artifacts. In the near future, AI applied to cardiac imaging will likely allow for automated diagnosis of disease states. Machine learning, a subset of AI, goes one step farther. Rather than programming the code by which a computer can interpret data or accomplish a task, machine learning allows the computer algorithm to continually improve its interpretations by exposure to more data. Simply stated, the machine "learns" and gets better over time without the need to be explicitly programmed by a human. Deep learning is a subset of machine learning in which the human is further removed from data analysis. Deep learning models use artificial neural networks (that mimic human/biologic neurons) to interpret data. Unlike machine learning algorithms, deep learning allows for data interpretation and self-directed analysis without the need for ongoing human programming.^1-7

Historically, AI, machine learning, and deep learning algorithms have been applied to static images in radiology. "Computer vision" techniques have been successfully used to detect suspicious lesions for breast cancer on mammograms and in chest radiographs for lung cancer.^8,9 The dynamic nature of cardiac imaging modalities (particularly cine imagine) is a particular challenge for AI compared with analysis of static images. Many companies are recently addressing this challenge, thereby enhancing the enthusiasm for AI among the cardiac imaging community. Presently, AI-based software that aids in image measurements is now commercially available for echocardiography, cardiac magnetic resonance imaging, cardiac computed tomography, and nuclear cardiology. Because echocardiography remains the cornerstone imaging modality in cardiology, this article predominantly focuses on applications of AI within the realm of echocardiography.

As the number of patients with cardiovascular disease continues to grow (along with the increasing complexity of patients with cardiovascular disease), it is likely that the number of echocardiographic studies will also experience parallel growth. The predicted supply-demand mismatch between growing patient volumes and the number of cardiologists available to interpret studies may potentially be addressed by AI.¹⁰

Indeed, the ability for a program to automatically perform chamber measurements of two-dimensional (2D) echocardiograms is now a reality. In the future, this may greatly impact echocardiography throughput and efficiency. Furthermore, due to variable image quality, sonographer/cardiologist experience, and challenging boundary detection, there is substantial variability with the manual quantification of echocardiographic parameters. The use of AI brings the hope of reducing variability and improving accuracy. Multiple companies/vendors and even investigators have demonstrated the ability to successfully perform automated echocardiographic recognition and interpretation of common 2D and three-dimensional (3D) structures and parameters and disease states.^11-17

Early forays into AI echocardiographic interpretation has focused on 2D echocardiography. In a simplified workflow, a full transthoracic echocardiogram DICOM is loaded into an AI-based computer program, and in less than 1 minute, the computer is able to identify the echocardiographic views and automatically measure a variety of 2D parameters. These parameters include left ventricular (LV) dimensions/hypertrophy, end-diastolic/end-systolic volumes to determine LV ejection fraction (EF), and atrial volumes. Additionally, automated detection of Doppler parameters (E, A, e', TR velocities) coupled with atrial volumes and LVEF could be used to determine diastolic function. Furthermore, others have shown that AI-based algorithms can automatically evaluate the severity of aortic stenosis (i.e., aortic valve area) either by evaluating aortic valve images or using quantitative parameters. Moreover, automated detection of 2D LV global longitudinal strain is already a reality.¹⁶ In the future, it is conceivable that before a cardiologist even opens a clinical echocardiogram, the vast majority of the quantitation will already have taken place automatically.

Beyond 2D echocardiography, the application of AI to 3D echocardiography is promising. 3D echocardiography is one of the most important developments in cardiac imaging.¹⁸ Left and right heart-chamber volumes derived from 3D echocardiography have been shown to correlate better with the reference standard of cardiac magnetic resonance imaging compared with 2D assessments. The most recent American Society of Echocardiography and European Association of Cardiovascular Imaging guidelines recommend the use of 3D echocardiography.¹⁹ Despite this recommendation, 3D echocardiography has not been fully adopted into clinical practice because it requires special training and expertise. The recent development of machine learning algorithms has resulted in the automated measurement of LV and left atrial volumes not only at their maximum and minimum values but also dynamically throughout the cardiac cycle.^13,20,21 Further refinements to this machine learning algorithm have also recently allowed for the determination of right ventricular volumes. This machine learning technique takes about 1 minute on average to provide fully automated analysis and requires little manual editing of the endocardial borders. Utilizing AI for 3D echocardiography may be a pivotal moment in echocardiography as more cardiologists incorporate this important technology into their clinically busy laboratories.

In addition to automated quantification of 2D and 3D echocardiograms, AI has the potential to revolutionize the way we read echocardiograms. Standard echocardiographic acquisitions are based on sonographer-obtained views, often starting with the parasternal views and followed by apical and finally subcostal views. Presently, the interpreter must integrate acquisitions from multiple non-consecutively acquired views in order to arrive at a conclusion or diagnosis. For instance, in order to determine the degree of aortic stenosis by continuity equation, the LV outflow tract diameter (parasternal long axis) must be integrated with Doppler tracings from the LV outflow tract and aortic valve (apical 3/5 chamber and occasionally suprasternal views) while also paying attention to LVEF (apical 2/4 chamber to obtain a biplane EF). Ongoing advancements in AI now allow automated classification of each acquisition image within a DICOM dataset. Theoretically, when one queries aortic valve area assessment, all the relevant images will be displayed together despite not being acquired consecutively. The same applies to parameters of LVEF, diastolic function, or valvular regurgitation. The underlying principle of this relies on the ability of the computer to recognize each captured image and categorize the images according to clinical utility. In the future, we will be able to read echocardiograms according to the particular chamber or valve being evaluated instead of the sequence in which images were acquired.

Going one step farther, AI may help improve echocardiography laboratory workflow and efficiency. By automatically analyzing key parameters, future programs should enable us to "triage" a list of unread echocardiograms by the probability of being normal or having abnormal findings. Accordingly, the reader will be able to prioritize interpreting abnormal studies first, leaving normal studies for later in the day. Similarly, the ability for AI programs to even detect disease states (such as hypertrophic cardiomyopathy, amyloidosis, or pulmonary hypertension) when studies meet specific echocardiographic criteria is also promising.¹¹

Yet another area where AI is having a positive impact on echocardiography is the image-acquisition process itself. Using a deep-learning approach, new software programs combine real-time image quality assessment with adaptive anatomic guidance to allow individuals with limited echocardiography training to acquire standard echocardiography images. This new paradigm shift in the application of AI may facilitate access to echocardiograms in resource-limited settings and in clinical situations where immediate interrogation of cardiac structure and function is needed.

One weighty question many have is whether echocardiographers will be replaced by computers in the future. If we observe our colleagues in radiology who have been using AI tools in their practice to aid in image interpretation, we shouldn't be concerned (yet). The present state of AI when applied to echocardiography is impressive, but AI still cannot integrate and extrapolate from a patient's clinical data, complementary diagnostic studies, or an echocardiographer's own prior experience when arriving at a conclusion. The human mind is still far more complex than AI, and it will likely be far into the future before we can rely solely on AI to interpret cardiac imaging studies. With that noted, it's fair to believe that an echocardiographer who uses AI will be more accurate, faster and efficient compared with one who does not.^22-25

References

1. Deo RC. Machine Learning in Medicine. Circulation 2015;132:1920-30.
2. Dilsizian ME, Siegel EL. Machine Meets Biology: a Primer on Artificial Intelligence in Cardiology and Cardiac Imaging. Curr Cardiol Rep 2018;20:139.
3. Gandhi S, Mosleh W, Shen J, Chow CM. Automation, machine learning, and artificial intelligence in echocardiography: A brave new world. Echocardiography 2018;35:1402-18.
4. Henglin M, Stein G, Hushcha PV, Snoek J, Wiltschko AB, Cheng S. Machine Learning Approaches in Cardiovascular Imaging. Circ Cardiovasc Imaging 2017;10:e005614.
5. Alsharqi M, Woodward WJ, Mumith JA, Markham DC, Upton R, Leeson P. Artificial intelligence and echocardiography. Echo Res Pract 2018;5:R115-R125.
6. Sengupta PP. Intelligent platforms for disease assessment: novel approaches in functional echocardiography. JACC Cardiovasc Imaging 2013;6:1206-11.
7. Johnson KW, Torres Soto J, Glicksberg BS, et al. Artificial Intelligence in Cardiology. J Am Coll Cardiol 2018;71:2668-79.
8. Yassin NIR, Omran S, El Houby EMF, Allam H. Machine learning techniques for breast cancer computer aided diagnosis using different image modalities: A systematic review. Comput Methods Programs Biomed 2018;156:25-45.
9. Qin C, Yao D, Shi Y, Song Z. Computer-aided detection in chest radiography based on artificial intelligence: a survey. Biomed Eng Online 2018;17:113.
10. Narang A, Sinha SS, Rajagopalan B, et al. The Supply and Demand of the Cardiovascular Workforce: Striking the Right Balance. J Am Coll Cardiol 2016;68:1680-9.
11. Zhang J, Gajjala S, Agrawal P, et al. Fully Automated Echocardiogram Interpretation in Clinical Practice. Circulation 2018;138:1623-35.
12. Madani A, Arnaout R, Mofrad M, Arnaout R. Fast and accurate view classification of echocardiograms using deep learning. NPJ Digit Med 2018;1:6.
13. Narang A, Mor-Avi V, Prado A, et al. Machine learning based automated dynamic quantification of left heart chamber volumes. Eur Heart J Cardiovasc Imaging 2018;Oct 9:[Epub ahead of print].
14. Volpato V, Mor-Avi V, Narang A, et al. Automated, machine learning-based, 3D echocardiographic quantification of left ventricular mass. Echocardiography 2019;36:312-9.
15. Leclerc S, Smistad E, Pedrosa J, et al. Deep Learning for Segmentation using an Open Large-Scale Dataset in 2D Echocardiography. IEEE Trans Med Imaging 2019;Feb 22:[Epub ahead of print].
16. Knackstedt C, Bekkers SC, Schummers G, et al. Fully Automated Versus Standard Tracking of Left Ventricular Ejection Fraction and Longitudinal Strain: The FAST-EFs Multicenter Study. J Am Coll Cardiol 2015;66:1456-66.
17. Narula S, Shameer K, Salem Omar AM, Dudley JT, Sengupta PP. Machine-Learning Algorithms to Automate Morphological and Functional Assessments in 2D Echocardiography. J Am Coll Cardiol 2016;68:2287-95.
18. Lang RM, Addetia K, Narang A, Mor-Avi V. 3-Dimensional Echocardiography: Latest Developments and Future Directions. JACC Cardiovasc Imaging 2018;11:1854-78.
19. Lang RM, Badano LP, Mor-Avi V, et al. Recommendations for cardiac chamber quantification by echocardiography in adults: an update from the American Society of Echocardiography and the European Association of Cardiovascular Imaging. J Am Soc Echocardiogr 2015;28:1-39.e14.
20. Tsang W, Salgo IS, Medvedofsky D, et al. Transthoracic 3D Echocardiographic Left Heart Chamber Quantification Using an Automated Adaptive Analytics Algorithm. JACC Cardiovasc Imaging 2016;9:769-82.
21. Medvedofsky D, Mor-Avi V, Amzulescu M, et al. Three-dimensional echocardiographic quantification of the left-heart chambers using an automated adaptive analytics algorithm: multicentre validation study. Eur Heart J Cardiovasc Imaging 2018;19:47-58.
22. Sengupta PP, Adjeroh DA. Will Artificial Intelligence Replace the Human Echocardiographer? Circulation 2018;138:1639-42.
23. D'hooge J, Fraser AG. Learning About Machine Learning to Create a Self-Driving Echocardiographic Laboratory. Circulation 2018;138:1636-8.
24. Chan S, Siegel EL. Will machine learning end the viability of radiology as a thriving medical specialty? Br J Radiol 2019;92:20180416.
25. Tajik AJ. Machine Learning for Echocardiographic Imaging: Embarking on Another Incredible Journey. J Am Coll Cardiol 2016;68:2296-8.

Keywords: Cardiac Imaging Techniques, Diagnostic Imaging, Aortic Valve, Artifacts, Patient Selection, Research Personnel, Stroke Volume, Atrial Fibrillation, Echocardiography, Echocardiography, Three-Dimensional, Algorithms, Artificial Intelligence, Aortic Valve Stenosis, Magnetic Resonance Imaging, Software, Tomography, Hypertrophy

< Back to Listings