Blog November 30, 2020

Emerging Applications of Artificial Intelligence in Cardiology

Artificial intelligence (AI) refers to machine learning based on deep neural networks that enable sophisticated computer recognition of subtle patterns within many layers of data. AI has been employed in medicine to identify pathologic specimens, automatically detect suspicious lesions in mammograms, more accurately read radiologic images, and identify retinal pathology with greater accuracy than trained ophthalmologists. Now, uses of machine learning in cardiovascular medicine are rapidly emerging, with the potential to expand physicians’ ability to make more accurate and prompt diagnoses and improve healthcare delivery. 

Applications already beginning to show benefit include the quicker assessment and treatment of strokes, faster and more consistent readings in nuclear imaging and other diagnostic radiology, and most recently, the earlier detection and monitoring of heart disease by applying AI to electrocardiogram (ECG) and ultrasound readings. 

This last area of development has seen particularly promising applications enter the marketplace. Early this year, the FDA cleared several AI-enhanced ultrasound technologies from Caption Health and EKO for use in evaluating COVID-19 infected patients for potential heart damage. In mid-October, GE Healthcare received FDA clearance for an AI-powered cardiovascular ultrasound system designed to automate and standardize echocardiogram (ECG) exams. The Ultra Edition package for GE’s line of Vivid ultrasounds incorporates machine learning algorithms that automatically detect the points in a 2-D image needed to measure the size of the heart’s left ventricle, an important metric for diagnosing heart failure and disease. The AI-enhanced system also semi-automatically measures blood flow and velocity, and records the transducer angle used by the ultrasound technician for each image, making it easier for later image review, with less variability. Combined, the new capabilities are expected to save 7-10 minutes per scan. Additionally, the system enables health care providers to reproduce a patient’s previous scans over time, to better monitor disease progression. 

In Europe, Cardiologs Technologies SAS announced a partnership with MicroPort CRM to distribute its AI-powered ECG analysis system to aid screening for arrhythmias such as atrial fibrillation (AFib) using ambulatory ECG recordings, marking the launch of the company’s EU business operations. MicroPort CRM is a leading developer and distributor of implantable devices for the treatment of heart rhythm disorders in France.

AFib affects about 33 million patients worldwide and is associated with an increased risk of severe stroke, heart failure, and death. It is often asymptomatic, with stroke as the first manifestation, and is associated with about a third of all ischemic strokes. Thus, experts believe that the ability to screen patients for AFib and quickly administer anti-coagulant treatment to those affected may reduce the healthcare burden of stroke.

Current state-of-the-art methods of detecting AFib have a relatively low positive predictive value of less than 59%, resulting in false positive misdiagnoses that cause patients stress and add significant cost to the healthcare system. In contrast, clinical data submitted by Cardiologs in support of its 2017 FDA clearance showed a positive predictive value of 91%, and a sensitivity (percentage of positive cases identified) of 97%. To develop its device, the Cardiologs team trained a neural network using over 500,000 (and growing) ECG recordings to recognize subtle patterns in cardiac signals for AFib, in a similar manner to expert cardiologists. 

Applications of AI to ECG readings have been the focus of other research, including at the Mayo Clinic, which has a database of over 7 million ECGs. The ability to accurately mine such data offers the potential to accurately predict heart failure noninvasively, inexpensively and within seconds. Now research is ongoing to use such readings to predict risk early in embolic stroke, and the be able to monitor patients and detect arrhythmias using AI compatible smartphones, and smart clothing and other consumer wearables.  

AI also offers the potential in cardiology to help personalize the complex treatment of patients with other co-morbidities who are taking numerous drugs from different doctors that might have deleterious interactions. Deep learning could help sort through a patient’s data, prior and current lab results, radiology reports, ECGs and procedure reports and offer suggestions for best managing that individual’s care based on current guidelines or data from large clinical studies. 

Some concerns remain, however, despite their promise.  The American College of Cardiology (ACC) states there is a need for greater clinical evidence that the new AI applications can truly help patient care, citing earlier generations of AI-enhanced technology (such as that used to automate the reading of certain mammographic scans) that ultimately did not live up to their promise. They also say that the way in which cardiology procedures are currently reimbursed needs to change if the investment in AI is to be justified. 

Could AI-enabled cardiology biomarkers emerge thanks to these new technologies? How will payers react to these new procedures? To what extent might these new advances help limit spending in cardiology? We’ll be watching this space.