Women’s bodies can be mysterious things—even to the women who inhabit them. But a wearable gadget called the Yono aims to replace mystery with knowledge derived from statistics, big data, and machine learning.
A woman who is trying to get pregnant may spend months tracking her ovulation cycle, often making a daily log of biological signals to determine her few days of fertility. While a plethora of apps promise to help, several studies have questioned these apps’ accuracy and efficacy.
Meanwhile, a woman who is trying to avoid pregnancy by the “fertility awareness method” may well not avoid it, since the method is only 75 percent effective.
The wearable from Yono Labs, a startup based in Silicon Valley, aims to help with both those dilemmas. Its earbud-like device is part of a new class of “hearables” that nestle inside the ear to pick up signals from the body.
Vanessa Xi, founder and CEO of Yono Labs, explains that her product came about through “personal painful experience.” The typical method of fertility tracking relies on a metric called basal body temperature (BBT), the lowest temperature the body attains during rest. If a woman is carefully measuring her BBT every day, she can spot the small temperature rise caused by the surge of hormones associated with ovulation.
But plotting BBT over the course of a month can be frustrating and imprecise, says Xi. “When I was trying to get pregnant, I had to wake up at the same time every morning and take my temperature immediately, before moving or getting out of bed,” she says. “It was really burdensome, and the data was everywhere. I wanted to solve this problem.”
Photo: Yono Labs
The Yono replaces that exacting regimen, which results in only one daily data point, with a stream of temperature data collected throughout the night. The user tucks the Yono earbud into her ear before going to sleep; the next morning, the Yono app sorts through the 70 to 120 temperature readings to determine her BBT.
The Yono is already on the market: Xi says her team released a beta version for testing in 2017 and has been steadily improving the device since then. Currently, the app uses the BBT data to plot a monthly fertility chart for the user.
The next step, she says, is to make accurate and personalized predictions about fertility. But although the Yono provides a lot of data to work with, that data isn’t clean and tidy. Temperature readings can be affected by head movements and sleep positions, and chunks of data may be missing if the earbud falls out or if the user forgets to put it in entirely.
That’s where Peter Song, a professor of biostatistics at the University of Michigan, comes in. After being introduced to Xi by one of his former PhD students, Song took the “very noisy data” generated by the Yono device and set out to create a “well-carved model that can predict with high precision the timing of ovulation,” he tells Spectrum.
In a paper published in the journal IEEE Transactions on Biomedical Engineering, Song describes the algorithms his team created to find the most relevant data points in the stream. They started with data cleaning and normalization to remove outlier readings. “For example, we got rid of data below 32 degrees Celsius, because that’s not biologically possible,” he says. Such low readings indicate that the device is positioned incorrectly or has fallen out, and that it’s really measuring the room temperature.
They also incorporated knowledge about the average woman’s ovulation cycle. “We know that 14 days after the beginning of the cycle is the most probable day of ovulation,” Song says. “So we incorporated that knowledge into our model to overcome measurement bias.”
Once the algorithm could identify a handful of very good data points from each night, Song used a type of statistical model called a hidden Markov model to determine the probabilistic relationship between the data points.
Xi plans to incorporate Song’s system into the Yono soon. In the first month of use, the Yono’s predictions will still rely on population statistics, but it will use data from each new cycle to make the predictions more personalized and precise.
Song says his work with the Yono team shows the potential for wearable devices that collect streams of data. “We want to put the data to better use,” he says, “so we can build a little more intelligence into these devices.”