Two of the primary vital signs are breathing and pulse rate. There are various solutions to monitor them, however, all require additional equipment and expertise to use. Smartphones are nowadays at almost every person's arm length, therefore, it could be cost-efective for crowd and personal health screening. Diaphragmatic breathing can be measured with Inertial Measurement Units (IMU). To optimize the breathing detection the smartphone has to be placed in the middle of the epigastric region. The tissue in the region vibrates because of the presence of the abdominal aorta which is also picked up by the IMU. Breathing, which is usually under 1 Hz during sleep can be ifltered out with a Bandpass filter. The heartbeat is present as vibrations which can be seen between 1-30 Hz. After filtering, breathing is detectable by a peak detector algorithm and can be diferentiated from noises.
Right now it is a growing need and supply for personal health devices and
applications. Companies are emerging in the telemedicine market and established
companies are creating divisions for both telemedicine and to develop commercial
devices [
Extensive review was done by [
A review was done by [
As a data recorder, we used an iPhone6. According to iFixit’s teardown an InvenSense MP67B 6-axis Gyroscope and Accelerometer Combo was found.(we did not receive any funding from Apple or iFixit) It is important to mention that the phone’s weight, morphology and the place where the IMU is to be found also matters. The weight matters particularly because the pulse vibrations have to vibrate the phone.
To acquire the data from the phone we were using Bernd Thomas’s iPhone application called SensorLog. Here we can partially choose which type of data we want to record, few of them are mandatory. We chose: “Accelerometer”, “Gyro” and “Altimeter”. Every sample is a row, which gets a timestamp. The sampling rate is configurable between 1-100 sample/second, we set it to 100. The high sampling rate is needed because of the fast vibration from which the pulse wave is calculated. In the application, data can be saved into a comma-separated values file (CSV) or JavaScript Object Notation (JSON) format. After recording, the file (measured values) was transmitted to the computer (server) by AirDrop.
We looked for the breathing signal on the accelerometer and gyroscope on the
accelerometer during ideal supine position and near-perfect placement (Figure 1).
It is crucial that in this position the phone measures abdominal breathing
movement, rather than thorax breathing movements. During Rapid Eye Movement
sleep(REM) one relies on abdominal breathing[
In comparison, the gyroscope is only afected by body morphology which can
vary based on posture. Yaw angles can not be calculated from just the
accelerometer, only roll and pitch angles [
We show the presence of respiration on the accelerometer, however we extract it
only from the gyroscope. We also show the pulse waveform on the gyroscope. The
ifgures have been created with matplotlib [
We extract ofline the pulse and abdominal breathing movement from the gyroscope’s angular velocities due to inherent problems with the accelerometer.
We are free to use Finite Impulse Response (FIR) and Infinite Impulse
Response (IIR) filters because we process the data ofline and also in this scenario
signal distortion is acceptable. For the former integer arithmetic is enough, but
latter requires floating-point arithmetic. Generally, a lower order IIR can achieve
similar results to a higher-order FIR filter, therefore we chose IIR Butterworth
bandpass filters. [
To produce a wave for heartbeat detection we are using an IIR Butterworth
bandpass filter with a low-pass cut-of frequency of 10 Hz and high-pass cut-of
frequency of 22 Hz with an order of 15. Again we applied the filter forward then
backward. Some post-processing is needed to produce a wave which could be
suitable to detect heartbeats. Both filters were created with Scipy using
Matlabstyle design[
During a regulated state like sleep, the detection of normal breathing becomes
much simpler. Usually, after an inspiration soon comes an expiration. We could
ifnd the inspiration and expiration peaks if they were significantly bigger than the
interference caused by the pulse. The condition that the inspiration and
expiration peak(which is negative) have to be close to each other easily handles noises
however it misses breaths without expiration and doesn’t handle most of the
abnormal breathing patterns [
In this paper, we showed that a smartphone, in this particular case an iPhone6 can be used to obtain a signal which approximates abdominal breathing movement and for detecting heartbeats. Measuring physiological signals just with a phone could provide data about people’s health and sleep at a scale and level which is unprecedented. We are confident that the following metrics can be calculated:
We suspect that many more features can be calculated, however that is less clear how robust would they be. At least for short time windows respiratory efort approximation could be highly useful to detect obstructive apnea. It is very important that IMU’s gyroscope gives more reliable, stable and robust data. Accelerometers output is dependant on the absolute orientation they are in.
1. We are not evaluating IMUs. We plan to create a simulation which points to the required IMU properties, with the appropriate simulation intra- and inter-user variability can be assessed. 2. We are focusing only on sleep because there is little movement to be found.
It gives the opportunity to extract more precise data about breathing and pulse rate than in any other scenario. 3. We did not validate against a reference device.
4. We measured our abdominal breathing movement and pulse rate. 3.2. Suggested and future work 1. Validating should be done against a reference device. 2. A further developed technique could estimate respiratory efort. 3. Frequency components and amplitudes of breathing and the pulse wave can highly overlap and change, therefore adaptive filtering is needed. 4. For the same reason and also for respiration efort estimation a more sophisticated adaptive peak detection is needed. 5. Calibration would be mandatory in an application which can have clinical significance. 6. Aces fusion should be developed for handling orientation deviations caused by displacement and body morphology. 7. Review and investigate sensor fusion options to provide robust and better data. Cimrman, R., Henriksen, I., Quintero, E. A., Harris, C. R., Archibald, A. M., Ribeiro, A. H., Pedregosa, F., van Mulbregt, P., and Contributors, S. . . SciPy 1.0–Fundamental Algorithms for Scientific Computing in Python. arXiv e-prints (Jul 2019), arXiv:1907.10121.