Using transdermal optical imaging technology, a smartphone camera can be used to extract information about the blood flow in the face. With good results in terms of accuracy.
Nuralogix is a Canadian Artificial Intelligence company that has patented an innovative technology (Transdermal Optical Imaging – TOI) capable of measuring several vital parameters by detecting blood flow in the face.
The technology is based on the fact that human skin is translucent. Light and its respective wavelengths are reflected in different layers under the skin and can be used to reveal information about the blood flow in the human face. This information can be captured by video images from conventional cameras.
Nuralogix has developed an app for iOS and Android smartphones, called Anura, which allows you to measure various health indicators in just 30 seconds by framing your face.
Anura measures heart rate, irregular heartbeats, breathing rate, blood pressure, heart rate variability, stress level and cardiovascular disease risk.
The measurement starts, after registering and defining your profile, by placing the smartphone in front of you and framing your face with the camera (the technology also works with those of laptops and tablets).
The software automatically detects and tracks the face, identifying key regions of interest (ROI). By holding the smartphone still for 30 seconds, the extracted facial blood flow data is sent to the cloud where the DeepAffex AI engine applies advanced signal processing and Deep Learning models developed by Nuralogix to predict physiological and psychological effects.
The processed results are then sent to the device for display and further analysis.
The software, which is not certified as a medical device, has been evaluated in a number of scientific studies that can be consulted on the manufacturer’s website.
These include a study published in the American Heart Association’s (AHA) journal Circulation: Cardiovascular Imaging, involving 1328 normotensive adults. The researchers used an advanced machine learning algorithm to create computational models predicting systolic, diastolic and heart rate pressure from facial blood flow data. The researchers used 70% of the dataset to train the models, 15% to test them, and the remaining 15% of the sample was used to validate model performance.
The results indicated that the models were able to predict blood pressure with a measurement bias±SD of 0.39±7.30 mm Hg for systolic pressure, -0.20±6.00 mm Hg for diastolic pressure.
In particular, the models were able to measure blood pressure with an average accuracy of 95% in subjects with blood pressure values of 100 – 139 for systolic, 60 – 89 for diastolic.
The app piqued my interest and I decided to install it on my iPhone and test it against the values measured by a sphygmomanometer with a cuff.
Here are my first impressions from some comparative measurements.
Registration is simple and can also be done with Apple credentials, deciding not to disclose your email. Defining your profile requires you to set some parameters (age, weight, height) and answer a few simple questions.
Once you have finished recording, the app starts up and a circle appears within which you have to frame your face in. After 3 seconds, if the framing is successful, the measurement starts and lasts 30 seconds. It’s very important to keep your hand still, otherwise the measurement gets stuck and you have to start all over again.
Personally, I found it more comfortable to sit down and rest my elbow on the table, also to compare more correctly the measurement made later with the sphygmomanometer.
When the measurement is complete, the software sends the data to the cloud for processing and in a couple of seconds you get your answer.
The data is divided into six categories: vital parameters, psychological state, mental state, physical state, risks and overall assessment.
In the vital parameters there are heart rate, respiratory rate (which in my case the software was
Psychological state is based on heart rate variability and cardiac load, which is calculated using the formula ‘heart rate x systolic pressure’. Mental state is measured by a stress index from 1 to 5 also based on heart rate variability.
Physical state depends on body mass index, age of facial skin, height to waist ratio and body shape index.
Detected risks include cardiovascular disorders, heart attack and stroke.
The general wellbeing index combines all the above values on a scale from 0 to 100.
Using the software is really simple. There is a simple but comprehensive explanation for each value or indicator. Comparing the readings with the sphygmomanometer, I verified that the diastolic pressure and heart rate values were almost the same as those measured with the medical device. The systolic blood pressure measured by Anura was on average 4-7 mmHg lower than that of the medical device in the range indicated by the clinical study. At values above 140 mmHg the measurement becomes less accurate, with a greater tendency to under-measure.
To work well, the light needs to be evenly distributed over the face and the smartphone needs to be held still. Anura will not work well if the light is too bright or dark or uneven on the face or if the face is not facing the camera. Hats and glasses can affect the reading.
Overall, I was positively impressed with the app which exceeded my expectations. It does not replace a medical device but can be useful to have a better awareness of your health.
Give it a try and send me your comments.
Anura is not the only app that detects vital signs from the face. The Israeli company Binah has developed its own technology similar to Nuralogix that can also measure oxygen saturation in the blood, but not blood pressure at the moment.
Unfortunately, the app is not available to the public but only to people who are supervised by a team that has a licence to use the software.
I have been testing the app for a few weeks and found that the heart and respiratory rate measurements are accurate, while the SpO2 reading seemed unreliable and too dependent on light conditions. The values I obtained, even after a short period of time, were very variable and in some cases absolutely far from those measured with a pulse oximeter.