The heart is made up of valves that open and close with blood flow as the heart muscle relaxes and contracts. Valvular heart disease (VHD) can have many underlying causes and occurs when these valves fail to work effectively. It is estimated 2.5% of the current United States population has VHD, and it is responsible for over 25,000 deaths each year. Due to advances in diagnostic technology, such as echocardiography, and advanced treatment approaches such as valve replacement, VHD is treatable in many cases. However, it is estimated that many patients go undiagnosed, despite millions of echocardiograms ordered in North America each year. Many human limitations still exist, particularly when it comes to cardiac image acquisition, interpretation, disease state identification, and prognostication. Artificial intelligence (AI) can assist in mitigating many of these limitations given its ability to drive complex-decision making in a short amount of time.
The effective acquisition of an echocardiogram requires significant training and experience. It is estimated that only about 83% of tricuspid valve images obtained by novice practitioners are evaluable. AI can assist by developing programs that guide image acquisition with a higher degree of accuracy by precisely identifying views that allow for the most valuable information. Deep learning can also assist with disease state identification, given its ability to recognize subtleties in data that humans cannot. A recently developed algorithm can automatically assess mitral valve regurgitation severity with over 99% accuracy by identifying “micro-pattern” information.
There are also many variables detected by echocardiography that are likely underutilized for disease state phenotyping because we don’t fully understand their value. AI can elucidate non-linear associations between many predictors and various outcome metrics without any prior knowledge. These abilities make it uniquely suited to uncover some of these less understood variables and guide future research and development.
The use of AI to manage VHD is not without drawbacks. Much of AI is not yet refined to reduce “noise” and other elements that limit its utility. There is also controversy regarding patient privacy, potential bias in programming, and data security, limiting its widespread adoption. However, given the ever-growing time constraints, patient data, and financial cost of poor health outcomes, improved AI models should be implemented stepwise to manage VHD more efficiently. AI tools can significantly improve diagnostic accuracy and pave new directions for research and development in this field and perhaps many others.
Sources: CDC, BMJ, Computers in Biology and Medicine
