The scene of doctor, patient and beeping machine is something we’re all familiar with. Let’s revise it a bit, and imagine a doctor remotely monitoring the vital signs of a chronic respiratory disease patient with wearable sensors. Our clinician sits in his office, reviewing the patient’s vital signs on his computer screen. Based on what he observes, he decides to call the patient into the clinic for a test.
Seems simple, but have you ever wondered what exactly those beeping machines are measuring, and how? There is a lot going on behind the scenes to make this situation possible in and outside the hospital. Let’s start with some key definitions.
The array of sensors in a Sensor Dot can measure various physiologic signals (waveforms) that can teach us something about health and disease when interpreted correctly. However, it takes some heavy-duty algorithmic (signal processing) work to get from these raw signals to medical insights that can be used by a doctor to make recommendations or treatment decisions.
Several steps are taken to extract relevant information from the signal. Some of the tools in the data scientist's toolbox include filtering to remove unwanted frequencies, artifact detection, signal fusion, and last-but-not-least feature detection. For the latter, physiologically meaningful (repeating) patterns are selected. All of these steps (and more) are needed to ensure that what is being detected is medically relevant. Once a set of features is extracted from the raw signal, they can be grouped into vital signs which tell us something about the body's vital functions (e.g. respiratory rate) and biomarkers. The latter is an important definition: in our case, a biomarker is any combination of features derived from physiologic waveforms that tells us something useful about a (patho)physiologic process.
At Byteflies, we have some very talented people building these end-to-end pipelines for signal processing. At this point, this may all still sound fairly abstract for the non-initiated. Continue reading for a more practical example!
Our organ systems — such as our nervous, respiratory, and circulatory systems — generate signals that can be measured using sensors. When measuring these signals often enough (at a high sampling frequency), a visual representation of the change in the physiologic signal over time is generated.
An electrocardiogram (ECG) signal visualizes the electrical activity generated in the heart muscle that makes it beat.
A photoplethysmogram (PPG) signal visualizes the pulsatile nature of blood flowing through vessels.
The waveforms discussed above contain a lot of information but are not easy to digest. Therefore, features are typically extracted from these waveforms. A feature is defined as a recurring shape, such as a peak or a valley, that tells us something physiologically interesting. Some features occur at unpredictable times (e.g. a sudden change in a motion sensor indicating the wearer is getting up from a chair), or they can be cyclical (e.g. in case of signals that are related to the heart beat).
Cyclical Feature Examples
From an ECG signal, a very large number of features can be extracted. The most common ones are the R-peak, denoting ventricular depolarization; the P-wave, denoting atrial depolarization; and the T-wave, denoting ventricular repolarization.
Similarly, a PPG signal contains a wealth of information. The most common ones are the systolic peak, and the minimum (valley) at the end of the diastolic phase.
Once a signal has been analyzed, it is capable of telling the doctor something about a patient’s health. But in many cases, analyzing multiple signals and the correlation between them will allow the doctor to make a much more informed decision. One of the strengths of the Byteflies platform is that this multi-modality is supported out-of-the-box. Not only can a single Sensor Dot record multiple synchronized signals at once, more than one Sensor Dot can be used simultaneously in different locations, thereby generating an even larger set of synchronized signals.
After all that introduction, let's go back to our patient suffering from chronic respiratory disease. Hundreds of millions of people suffer from some form of chronic respiratory disease1, in many cases accompanied by cardiovascular problems (so called comorbidities). In the clinical example, our doctor was observing a combination of vital signs and biomarkers from his patient: respiratory rate, heart rate, oxygen saturation, and activity. For all of these metrics, detailed minute-by-minute changes, as well as summary statistics over longer periods of time are available.
What? A biopotential sensor placed on the chest measures the electrical impulses within the heart that make it contract.
Insights? A very detailed window into cardiac function, including but not limited to heart rate, heart rate variability, autonomic regulation, arrhythmias, cardiac remodeling, and perfusion.
What? A bioimpedance sensor measures the change in conductivity across a small patch of skin. Sweat will increase the conductivity, generating a proxy for perspiration. EDA is also known as Galvanic Skin Response (GSR).
Insights? A useful and easy-to-measure proxy for activity, stress, and autonomic regulation if used together with other signals to provide context (e.g. high versus low activity).
What? A biopotential sensor placed on a muscle, usually a limb, to record the electrical impulses generated by contraction of skeletal muscles.
Insights? A measure of skeletal muscle and motor neuron function that can be very useful for diagnosing certain (neuro)muscular disorders.
What? An IMU actually consists of three separate sensors: 1) an accelerometer that records acceleration in three dimensions; 2) a gyroscope that records rotation around three axes; and 3) a magnetometer that records the strength and direction of the Earth's magnetic field.
Insights? Together, these sensors provide an accurate quantification of movement, (type of) activity, and even (relative) position. As such, an IMU is a very useful "add-on" sensor to generate context information for other signals (see for instance our respiratory disease case).
What? An optical sensor that emits light of certain wavelengths and detects light that is scattered back to a detector. The properties of the scattered light change with respect to changes in blood volume in the subcutaneous microvasculature (the small blood vessels under the skin).
Insights? In addition to cardiac parameters such as heart rate, heart rate variability, and autonomic regulation, PPG provides insights into hemodynamical regulation. In other words, the changes in the cardiovascular system that are related to regulation of blood pressure and vessel function. Finally, oxygen saturation, the ratio of oxygenated versus deoxygenated hemoglobin, can be derived.
What? A bioimpedance measurement on the chest will be influenced by chest wall movement and the volume of air in the lungs, and thus provide a proxy for respiration.
Insights? In addition to respiratory rate, more detailed analysis of the dynamics of breathing can be derived, such as labored breathing, tachypnea (shallow breathing), coughing ...