OR WAIT null SECS
With machine learning, USC researchers correlated measurements taken nonintrusively by an iPhone camera to detailed tonometry data.
Photo Credit: Ashleen Knutsen. Image has been cropped for size.
A lot of work is being done around Apple devices and heart disease—most of it involves Apple Watches and their built-in heart rate monitors. A new Scientific Reports study by researchers from the University of Southern California, however, found that a smartphone camera is able to provide just enough input to detect a key cardiac condition, when it’s paired with a carefully-validated algorithm.
Arterial stiffening can be a risk factor for heart disease, and it can lead to cardiac events and organ damage. Typically, it’s diagnosed with an MRI or a tonometry device, which measures pulse wave velocity (PWV) to determine how blood is circulating through the body. That machine can cost thousands of dollars.
The team from USC’s Viterbi School of Engineering used an iPhone camera to unobtrusively measure a single waveform from outside the skin.
"An uncalibrated, single waveform - that means that you eliminated two steps. That's how you go from an $18,000 tonometry device and intrusive procedure to an iPhone app," researcher Niema M. Pahlevan said in a USC release. Pahlevan worked with Peyman Tavallali and Marianne Razavi of Avicenza, a Los Angeles-based medical device company that’s building the app.
The team had published a previous study last year which showed that the iPhone camera was capable of detecting a single carotid pressure wave. In this exercise, they were looking to validate the algorithms built to translate that measurement into a PWV equivalent.
They validated their machine learning model using tonometry data from the Framingham Heart Study. With about 85% accuracy, they were able to correlate their own PWV measurements to data from thousands of patients in that historic study.
“In this paper, we have introduced a novel artificial intelligence method to estimate the carotid-femoral pulse wave velocity,” the team concluded. Their method, based on a newly-introduced method called Intrinsic Frequency, “uses only an uncalibrated carotid pressure waveform with typical clinical variables such as blood pressure.”
Spotting arterial stiffening without the need for electrocardiography or a femoral tonometry reading is what allows the researchers to predict cardiac risk using only the iPhone camera, but it’s the underlying machine learning models that make it all possible. Pahlevan was sure to note that in the study’s accompanying release.
“A lot of people have tried to bring machine learning to medical devices, but pure AI by itself doesn’t work,” he said. “When you get a high correlation, you can be missing all of the diseased patients because, in medicine, the outliers are the cases you want to capture—they’re the important ones.”