Artificial intelligence used to detect early-stage heart disease
Posted: 9 January 2019 | Nikki Withers (European Pharmaceutical Review) | No comments yet
Mayo study uses artificial intelligence to create inexpensive, widely available early detector of heart disease…
Study findings show that applying artificial intelligence (AI) to the electrocardiogram (ECG) enables early detection of left ventricular dysfunction and can identify individuals at increased risk for its development in the future.
The research, published in Nature Medicine, found that the accuracy of the AI/ECG compares favourably to other common screening tests such as prostate-specific antigen for prostate cancer and mammography for breast cancer.
“The ability to acquire an ubiquitous, easily accessible, inexpensive recording in 10 seconds – the ECG – and to digitally process it with AI to extract new information about previously hidden heart disease holds great promise for saving lives and improving health,” says Paul Friedman, senior author and chair of the Midwest Department of Cardiovascular Medicine at Mayo Clinic.
Asymptomatic left ventricular dysfunction (ALVD) is characterised by the presence of a weak heart pump with a risk of overt heart failure. It is present in three to six percent of the general population and is associated with reduced quality of life and longevity. However, it is treatable when found.
Currently, there is no inexpensive, noninvasive, painless screening tool for ALVD available for diagnostic use.
To address this, Paul Friedman and colleagues tested whether ALVD could be reliable detected in the ECG by a properly trained neural network.
The team used paired 12-lead ECG and echocardiogram data, including the left ventricular ejection fraction (a measure of contractile function), from 44,959 patients at the Mayo Clinic, and trained a convolutional neural network to identify patients with ventricular dysfunction, defined as ejection fraction less than 35 percent, using the ECG data alone.
When tested on an independent set of 52,870 patients, the network model yielded values for the area under the curve, sensitivity, specificity, and accuracy of 0.93, 86.3 percent, 85.7 percent, and 85.7 percent, respectively.
Furthermore, in patients without ventricular dysfunction, those with a positive AI screen were at four times the risk of developing future ventricular dysfunction compared with those with a negative screen.
“This suggests the network detected early, subclinical, metabolic or structural abnormalities that manifest in the ECG,” says Friedman.