• Electrocardiography
• Diagnostic Imaging

A hundred years ago, in 1924, Willem Einthoven won the Nobel Prize for Physiology and Medicine for his invention of electrocardiography. Over a century, we have continued to perfect the science behind ECGs, making them easier to acquire, improving their quality, and defining pathognomonic features of conditions spanning a wide range of rhythm and conduction disorders. Their clinical value, along with their easy and low-cost acquisition, has made them the most widely used test in contemporary cardiovascular practice across the world. The diagnostic process of interpreting ECGs includes pattern recognition, with differences in expertise posing a risk for diagnostic failures. Moreover, there may be subtle patterns that are currently undetectable or too complex for human comprehension. Both these challenges can be solved by artificial intelligence (AI).

Applications of AI, armed with a new generation of deep learning methods, have demonstrated an excellent ability to learn complex patterns across various data types. A majority of computational algorithms have thus far leveraged the rich voltage data acquired by ECG devices, often 300–500 voltage recordings of the cardiac electrical activity per second over 12 channels for 10 s or longer. Traditional computational methods, as well as the earliest AI applications, leveraged these voltage data to diagnose rhythm and conduction disorders, with the intended goal of assisting clinicians in their diagnostic process.

However, a major advance in the domain of AI for ECGs was subsequent deployments that could augment human capacity and identify signatures of cardiovascular diseases, which were previously considered to not be detectable on ECGs. These include the detection of structural heart disorders on ECGs, something that typically requires advanced cardiac diagnostic testing. There was also novel prognostic …