I Heart Radar

A contactless ECG device opens the door to continuous monitoring of the heart, and early detection of abnormalities.

Nick Bild
a month agoMachine Learning & AI
(📷: J. Chen et al.)

One of the most important physiological signals is the electrocardiogram (ECG), which provides vital information about the function of the heart and can be used to diagnose various diseases. Research has suggested that by continuously monitoring and analyzing ECG signals, many chronic cardiovascular diseases could be better diagnosed. Unfortunately, continuous monitoring is a challenge in most cases, because current approaches to capture ECG signals rely on electrodes being in physical contact with the surface of the skin. This is often seen as uncomfortable or burdensome, and under certain circumstances — such as those of burn patients or premature babies — not possible.

A new, non-invasive take on the ECG may make it a more realistic option for continuous monitoring — and may also open up ECG monitoring to classes of patients currently excluded. A research group at the University of Science and Technology of China have developed a contactless ECG monitor that makes use of millimeter wave radar and deep learning to infer the electrical signals generated by the heart by observing its mechanical activity.

The technique transmits 77 GHz millimeter-wave radar waves, with 4 GHz of bandwidth, towards the patient’s torso area, which are then modulated by cardiac motions. The modulated waves are then reflected back to a receiver. The data processing pipeline isolates motions coming from different parts of the torso to zero in on signals from the proper anatomical region. Next, micro-motions are amplified to both eliminate interference from stronger motions, such as breathing, and also to improve the signal-to-noise ratio of cardiac motions. Finally, a round of cardiac signal focusing and spatial filtering is conducted to reveal the isolated cardiac signals, reduce any remaining noise, and reveal the true cardiac motions.

To convert the cardiac motions into ECG signals, a hierarchical deep neural network with an encoder-decoder architecture is employed. This design is capable of leveraging the temporal-spatial complexity of radio frequency input for this purpose. After training this machine learning model, it was evaluated in a study consisting of 36 participants, four physiological conditions (normal breathing, irregular breathing, post-exercise, and sleep), and 200 experimental trials. Over ten hours of total radar measurements were captured, and as a ground truth reference point, corresponding ECG measurements were also recorded.

The contactless ECG was found to be able to infer the timing of the Q-peaks, R-peaks, S-peaks, and T-peaks with a median error of 14 milliseconds (ms), 3 ms, 8 ms, and 10 ms, respectively. Morphological accuracy was also assessed in this study, and the analysis showed the new method to achieve a median Pearson correlation of 90% and a median root mean square error of 0.081 millivolts compared to the ground truth measurements. Further analysis of the timing measurement accuracies showed that the system has the potential to detect and diagnose heart arrhythmias.

While the current prototype system is not yet ready for clinical use, the accuracy shown thus far demonstrates its potential with a bit further development. One step in that direction would be to evaluate the device on patients with heart problems — the studies have focused on healthy individuals to date. To facilitate this development, the researchers have publicly released their dataset linking radar measurements with ECG readings.

Nick Bild
R&D, creativity, and building the next big thing you never knew you wanted are my specialties.
Latest articles
Sponsored articles
Related articles
Latest articles
Read more
Related articles