Artificial Intelligence in Echocardiography

Echocardiography is a common and essential tool in the diagnosis of cardiovascular disease. Using ultrasound, this technique allows for non-invasive assessment of cardiac function, including systolic function, diastolic function, heart chamber quantification, and diagnosis and quantification of valvular abnormalities. Usage of echocardiography has increased each year; over 7 million echocardiograms were performed in 2011 in the Medicare Population alone. Cardiovascular disease remains the leading cause of death worldwide, and allowing for widespread usage of echocardiography could result in earlier diagnosis and treatment of cardiovascular disease and potential reduction in healthcare disparities.

The procedure of an echocardiogram first requires image acquisition which is then followed by image analysis and interpretation. Image acquisition is traditionally performed by cardiac sonographers (technicians) or physicians. Image review and interpretation is performed by specialist physicians, typically cardiologists or radiologists. Each step in the process historically requires a high level of training and specialized equipment which limits its use in under-resourced areas. However, given the high level of skill required to operate this equipment there remains a need for additional technologies to aid in the acquisition and interpretation of imaging. As ultrasound technology has improved, however, costs and size of equipment have decreased and use of bedside ultrasound to guide clinical decision making has become increasingly common. This bedside ultrasound is focused on specific questions, such as diagnosis of a pleural or pericardial effusion, and can be readily taught to non-experts.

One of the potential tools to overcome these limitations is artificial intelligence (AI). AI is the use of computer programs to mimic the cognitive function of the human mind in order to learn and solve problems. Echocardiology is a particularly ripe field that may benefit from the use of artificial intelligence. Due to anatomical differences and dynamic clinical situations, the process of image acquisition and analysis can vary widely between patients. Although simple computer algorithms fail to integrate these differences, artificial intelligence may allow for machine-assisted image acquisition and analysis. In the last several years, there have been numerous studies of computer-assisted analysis of echocardiography. The use of this technology may speed the acquisition of echocardiographic images, reduce the amount of training needed to acquire and analyze images, improve diagnostic quality, and reduce interobserver variability in the analysis of echocardiographic images.

By integrating artificial intelligence-assisted image acquisition and analysis with ultrasound technology, it may be possible for minimally trained operators in underserved areas to use echocardiography to accurately diagnose cardiovascular disease.

This study will be supervised by the Echocardiography Laboratory and the Internal Medicine Residency Program in the Department of Medicine at the NewYork Presbyterian Columbia University Medical Center. The study will take place on the medical resident inpatient cardiology ward services which are primarily housed in the Milstein Hospital 5 Garden South ward. The primary subjects of the study are the medical residents in the Department of Medicine who are rotating through the inpatient cardiology ward services.

During this study, the medical residents will undergo a training session introducing them to the basic concepts of echocardiography and the usage of either the Bay Labs echocardiology AI ultrasound system or a standard echocardiography system (a native Terason echocardiography machine). In this protocol, the echocardiographic images they obtain using these tools will be called a "study echocardiogram." This is in contrast to a "formal echocardiogram" that is performed by cardiac sonographers, cardiology fellows, or cardiology attendings and analyzed by cardiology attendings who specialize in echocardiography. The medical residents will use either the Bay Labs system or the native Terason system to perform echocardiograms on patients admitted to the cardiology ward teams who either have undergone or are planned to undergo a formal echocardiogram within 1 day of the study echocardiogram. The explicit goals of this project are (1) to determine if it is possible for the medical residents as novice users to acquire and interpret echocardiographic images and (2) whether the use of the Bay Labs system aids in education over the standard system. The Bay Labs echocardiogram that is performed as part of this study is not FDA approved to aid in clinical decision making, and as such the results of the study echocardiograms will not be used to change clinical management.

There are a total of 4 cardiology resident teams on the inpatient cardiology wards, and the teams are designated using a letter system: A, B, C, and D. Each of these teams is comprised of a first-year medical resident and a third-year medical resident. The A and C teams form one team dyad and the B and D teams form a second team dyad. The teams are covered at night by a separate team of medical residents. Each dyad is primarily supervised by a pair of cardiology attendings who oversee the care of the general cardiac patients. In addition to the service cardiology attendings, the teams may manage patients under the care of heart failure cardiologists or other private cardiologists. Each team can carry a total of 10 patients. The admitting structure of the teams allows for 7 new patients to be admitted to the cardiology ward services per 24 hour period. During the day, 3 patients can be admitted to the long-call team and 1 patient to the short-call team. At night, 3 patients can be admitted to the overnight (long-call) team.

Prior to the initiation of the study, a schedule will be created where each 4 week cardiology inpatient ward rotation block will be assigned to use Bay Labs technology or the native Terason system. At the start of their cardiology inpatient ward rotation, the medical residents will be approached to consent to take place in the study. The medical residents who consent to the study will undergo a pre-test asking them about their comfort with the performance of and interpretation of echocardiography. They will then undergo a training session introducing them to the basics of echocardiography, the study protocol, and the operation of either the Bay Labs echocardiography system or the native Terason echocardiography system. After receiving this training, they will be able to start performing echocardiograms in the study.

Each morning, each medical resident will able to designate between 1-3 patients on their service to undergo an echocardiogram. The inclusion and exclusion criteria are listed in the protocol. The patients will be approached by the an investigator or research assistant to consent them for the study. Participants will then undergo an echocardiogram performed by the medical residents and images will be stored for later analysis. After each echocardiogram, the medical residents will fill out a short form documenting the echocardiogram. Questions on this form will include demographics, clinical features, technical aspects about the test (such as whether specific views could be obtained), and ask basic questions about operator estimated cardiac function that was observed during the test.

At the end of the rotation, the participating residents will take a post-test on their comfort with the performance of and interpretation of echocardiography. They will be offered the opportunity to provide feedback on the study. The medical residents will be offered feedback on their echocardiograms during and after the study period to aid in learning.

After the end of the study period, the data will be analyzed. Results from the pre-test and post-test will be analyzed to determine if the medical residents had improvement in their knowledge of echocardiography during the rotation. The echocardiograms will be analyzed and compared with the formal echocardiograms by the study investigators and attending echocardiographers to determine if the designated image views could be obtained, the time needed to obtain the images, and the quality of the images in comparison to the formal echocardiogram. The left ventricular systolic function visually observed by the medical residents during the study echocardiogram, the autoEF calculated by the Bay Labs software in those patients in those groups, and the left ventricular ejection fraction from the formal echocardiogram will be compared.

This study is of no more than minimal risk to the patients enrolled. The test itself (a transthoracic echocardiogram) is noninvasive and has no radiation exposure. In addition, the medical teams will be prohibited from making clinical decisions based on the Bay Labs echocardiogram which will instead be based on the formal echocardiogram. At the end of the study, the data will be permanently de-identified.