Emerging Role of Artificial Intelligence in Echocardiography

Abstract

Artificial intelligence (AI) is revolutionizing health care. The sudden spurt in the growth of AI in healthcare is driven by increased accessibility to datasets and the availability of user-friendly software to input data. Techniques for processing data through multilayered networks have evolved in the last few years. Machine learning (ML) is a subfield of AI that allows computers to learn without specific programming.[1] ML can learn rules, identify patterns progressively from large datasets, and effectively predict and make intelligent decisions in many areas. ML has revolutionized the field of radiological imaging.[2] Cardiovascular imaging is the primary diagnostic contributing to understanding abnormalities of anatomical structures such as ventricles, valves, or coronaries and measuring various parameters, including ejection fraction, perfusion defect, shunts, blood flow dynamics, or the extent of coronary stenosis.[2] Echocardiography is the most commonly used imaging modality to guide perioperative diagnoses, evaluations, and surgical decision-making.[3] Trans-thoracic echocardiography (TTE) is a vital diagnostic tool, but trans-esophageal echocardiography (TEE) is a vital monitoring and real-time hemodynamic evaluation tool during most cardiac surgeries. Data science is making significant inroads in the field of cardiovascular imaging. AI has emerged as a powerful tool in echocardiography. AI has revolutionized the acquisition, analysis, and interpretation of cardiac ultrasound images. The critical application of AI in echocardiography is image enhancement. AI can enhance image quality, detect abnormalities, and automate flow velocities and distance measurements, usually done manually by physicians. AI algorithms can measure the heart walls’ thickness, the heart chambers’ size, and the ejection fraction.[4] AI algorithms can analyze large amounts of data and identify patterns that may not be immediately apparent. Its integration with the electronic health record and pathology can assist in diagnosis and comprehensive treatment planning.[5] AI algorithms can remove noise, correct distortions, and improve the overall quality of ultrasound images, making images more accessible and facilitating accurate diagnoses.[6] AI can reduce the risk of errors, bring efficiency to reduce costs, and improve the quality value of the data.[7] M-mode and Doppler echocardiography are commonly used for cardiac examination. M-mode has high temporal and spatial resolution and effectively captures subtle motion patterns. The Doppler can acquire a velocity-time image to assess valvular regurgitation and stenosis. ML can automate several measurements associated with M-mode and Doppler.[4] AI use has been validated for automated quantification of ventricular volumes or ejection fraction, global longitudinal strain, atrial size or function from 2D and 3D acquisitions, assessment of myocardial thickening, endocardial excursion, and regional wall motion abnormalities detection.[8] The cardiac anatomy is complex, and sonogram classifications vary. The heart needs to be comprehensively evaluated by 10–30 section scans. Physicians with insufficient experience may fail to provide accurate, standardized analysis as minor angle differences among sections may make image acquisition difficult.[4] Rapid standard section recognition with AI can shorten the evaluation time and enhance detection ability and accuracy. Madani et al.[9] found AI to have better echocardiographic recognition accuracy (97.8%) than physicians (70.2–83.5%). Although the use of AI in TTE is expanding frantically, the application of AI and ML to TEE still needs to be established. The automated evaluation of TEE data is limited as TEE images and dynamics are complex and unstructured and demonstrate significant heterogeneity in various views to evaluate cardiac structures. The utilization of AI in this TEE has been restricted by the complex multi-view format of echocardiography and the inescapable requirement for human intervention to acquire and interpret the image. TEE plays a significant role in managing complex cardiac pathology and high-risk surgical patients. Its use is today inescapable during surgery/interventions and helps the cardiac anesthesiologist guide the surgeon/interventionist. Additional value from TEE images via deep learning strategies is thus the way forward. Steffner et al.[10] recently attempted to train a convolutional neural network to classify eight standardized TEE views using annotated intra-operative and intra-procedural TEE images. This deep learning model accurately classified a wide range of clinical and echocardiographic characteristics in the TEE views, representing a broad spectrum of anatomic pathology. This was possibly the first attempt to use deep learning strategies for the automated evaluation of intraoperative imaging. This study has opened new vistas for AI in TEE. AI requires voluminous data banks to train the algorithm. Sub-optimal data can lead to impaired analysis.[11] AI can match and be better than conventional echocardiography in interpretation and analysis. More studies displaying improved outcomes are needed to introduce it into routine clinical practice. A concerted multidisciplinary effort with engineers, computer scientists, sonographers, and physicians will be necessary to introduce AI in clinical echocardiography.[12] AI has the potential to revolutionize the field of echocardiography, making it more accurate, efficient, reproducible, and effective. Interpreting AI-generated outcomes with the additional information obtained from other resources is essential. AI may not replace the cardiac sonologist soon, but the sonologists should try to understand and apply AI tools more extensively and validate their efficacy. As AI technology advances, we can expect to see even more innovative and groundbreaking applications of AI in echocardiography in the years to come.

Ähnliche Arbeiten