Summer Interning at Byteflies (2018 Edition)

Interested in an internship at Byteflies? Contact us!

Byteflies is always on the lookout for excellent summer internship candidates. Last summer, we had the pleasure of hosting Annabel Descamps and Nicolas Coucke in the Signal Processing & Data Science team at Byteflies and Case Western Reserve University (CWRU). Annabel worked on deriving oxygen saturation (SpO2) from reflective PPG signals and Nicolas worked on cardiorespiratory coupling, a term describing the interactions between the autonomic, respiratory and cardiovascular control systems. Read on to learn more!

Cardiorespiratory Coupling

by Nicolas Coucke

Try this: make sure you feel your heartbeat and then hold your breath. After a while, you should feel your heart slow down. This is because there are mechanisms in your brainstem that link your heartbeat and respiration. These mechanisms are collectively known as cardiorespiratory coupling (CRC).

Why is it important to measure CRC? Some diseases may disrupt these mechanisms in the brain, which may cause a change in CRC. This loss of coupling often happens before any other symptoms become apparent. So detecting a loss in CRC might give doctors an early warning that something is wrong.



CRC is usually divided into two types. The most obvious type is Respiratory Sinus Arrhythmia (RSA). For this type, your heart rate increases during inspiration and goes back down during expiration. Meaning that for RSA, respiration influences heart rate. The opposite happens for the second type, Cardioventilatory Coupling (CVC), where a heartbeat actually triggers the onset of a respiration cycle.

Two conditions that might cause a disruption in CRC are sepsis and schizophrenia. Sepsis can happen when the body’s own immune defenses overreact to an infection, causing inflammation in the whole body, which is often lethal. Since CRC is regulated in the brainstem which is very sensitive to inflammation, changes in CRC can be early indicators of sepsis. Similarly, certain brain abnormalities that cause the disturbed sense of reality in schizophrenia, can also affect the areas where CRC is regulated.

To quantify coupling, we need to measure both respiration and the heartbeat simultaneously. Traditionally, measuring these signals is done in hospitals with bulky and expensive equipment, which makes it harder for patients to be continuously monitored. This is where the Byteflies Sensor Dot comes into play as it can measure heartbeat (via electrocardiogram, ECG) and respiration (via bioimpedance) simultaneously and in a small wearable package. The main objective of my internship was to derive CRC measures from ECG and respiratory signals from the Sensor Dot, reference hospital equipment, and to compare the accuracy of the different devices.

The first week of the internship took place at the Byteflies office in Antwerp. I became familiar with the Sensor Dots and accompanying software. It was good to know that I was not on my own to do this project. After this week, I was prepared to start at CWRU in Cleveland, Ohio. Byteflies collaborates with this American university to validate and test their equipment. Over there, I could count on the excellent advice of professors from both the engineering and the medical faculties. We recruited some test subjects and hooked them up to the Sensor Dot and a Philips MP70 patient monitor*. We then did measurements while they performed specific breathing exercises.

*This is a typical ICU monitor used in intensive care units to continuously monitor a patient’s vital signs.

The Sensor Dot gave us an ECG and respiration (bioimpedance) signal. The clinical equipment returned an ECG and two respiratory waveforms: a capnogram* and bioimpedance signal. In the figure below, I show two respiratory waveforms: one recorded via the patient monitor (capnography, blue) and the other recorded with the Sensor Dot (bioimpedance, red).

*A capnograph measures the exhaled carbon dioxide concentration via a nasal cannula.


From these signals, I calculated measures of respiratory rate and coupling. Generally, the signals derived from both the Sensor Dot and the clinical device were very similar, indicating that it could be useful to monitor CRC without expensive or bulky hospital equipment*.

*Nicolas's results will be used in a future publication.

I want to thank everyone at Byteflies and CWRU for the time and effort they took to help me. This internship helped to develop my ability to approach challenging projects and motivated me to keep pushing the limits of medical technology. I would recommend it to everyone that is interested in data science!

Deriving Oxygen Saturation from Reflective PPG

by Annabel Descamps

Sleep Apnea is a disease that can cause your blood oxygen level to drop dangerously low. However, it's a disease many people are unaware of, given the lengthy and complicated process to access a diagnosis. Patients need to first get a polysomnography (PSG) through an overnight measurement at a hospital or specialised sleep center. This procedure is not easily accessible and consists of attaching a series of wires and devices to the patient.

psg (1)


As a contrast, through Byteflies’ Sensor Dots it is possible to monitor apneic events while sleeping, by allocating them on the chest. Therefore, for my internship I investigated their feasibility by focusing on oxygen saturation (SpO2).


ByteFlies_Perspective_OpenFirst, I spent a week at the Byteflies office in Antwerp to get familiar with the Sensor Dots and the team's work. After this training week, I flew to Cleveland, Ohio to work at CWRU, one of Byteflies’ collaborators. Once there, Nicolas (a fellow student) and myself had the opportunity to work with various professors, who often gave us advice and guidance on our projects. For my project, I used the theory of pulse oximetry which links features from the photplethysmography (PPG) signals to SpO2. It requires a detector and two LEDs of red and IR wavelengths. In the hospital, this is typically done with a fingerclip that uses transmissive PPG, where the detector is located on the opposite sides of the LEDs. The Sensor Dots however, are designed to be placed on more comfortable locations and therefore work reflective. This is a less established method for calculating SpO2 and my task was to figure out if the reflective PPG signals of the Sensor Dot could be used to derive a measure for the trend of the blood oxygen saturation (%SpO2).

Image Source:

To find out, we first constructed an experiment protocol that offered useful data for both of us, so we could combine our experiments. We managed to run our protocol on 9 subjects, including ourselves. To prove the Sensor Dot capabilities, I needed the subjects to lower their SpO2 to get a sufficiently large range of values. Therefore, three Sensor Dots were placed on the subjects forehead, arm and chest and were then asked to hold their breath twice with a resting period in between*.

*Annabel's results will be used in a future publication.

I am extremely glad I got to do this internship because I learned so much from it. I am also very grateful for the opportunity and constant feedback Byteflies gave me. Everyone in the team was always willing to answer my questions when needed. I now feel more prepared and confident to go into the real world and tackle other projects.