Current technologies allow us to learn more about what is happening inside our bodies. Different modalities have been used in recent years to extract different information concerning our health: Whether we are measuring the heart rate, blood pressure, blood glucose concentration, brain activity or using imaging modalities to get a better picture of our bones, eyes, muscles or nerves, we are constantly generating information about our bodies that can help physicians diagnose certain diseases or aid at treating patients.
A biomedical signal can be any kind of signal that is measured from the human body, examples include: 1-D signals, when we measure vital signals, 2-D or 3-D images of the organs, or audio signals in hearing aids. Usually, the obtained biomedical measurements need to be analyzed in order for the clinician to be able to infer information from them. For example, in EEG signals we commonly encounter motion artifacts that disrupt the signals and make it impossible to extract the necessary information. In Biomedical imaging, we often want to identify the different regions in an image using segmentation techniques and classify these to detect anomalies. Using methods of parameter estimation and models of the measurements, we can extract reliable information. Robust statistics play a big role in biomedical signal processing as real life data can often deviate from the model assumptions.
Many biomedical devices tend to be small or have to be affordable for the patient, which leads to very stringent conditions on power consumption and size of the devices. The challenge is often to design algorithms that comply with these constraints while showing high accuracy and reliability.
A project in cooperation with Roche Diagnostics GmbH, Mannheim, is on diabetes care; details can be found below.
In our Biolab (biological experiment laboratory), we have a selection of sensors to conduct measurements with, e.g., electrocardiogram (ECG), photoplethysmogram (PPG), electrodermal activity (EDA), respiration and so on. For the simulation of activities, we provide a cross trainer and will soon acquire an indoor cycle and a treadmill.
For more information on biomedical signal processing ,see the sections below or contact the respective Research Associates.
Diabetes Mellitus is a chronic disease that affects 347 million people worldwide and results in high blood sugar levels. It can lead to severe medical complications, such as eye and nerve diseases. An essential component of diabetes care, which can help delay or even prevent complications is careful self-monitoring of blood sugar levels by the patients. To this end, invasive blood sugar measurement devices, as seen in the image on the right, are used up to 4-5 times daily by the patient. These devices need to show high accuracy and good usability, while remaining cheap under low power constraints.
In a cooperation project with Roche Diagnostics GmbH in Mannheim, we develop a new generation of blood sugar measurement devices that require blood sample volumes in the nl-range (100 times less than state-of-the-art devices). Hereby, the induced pain from extracting a blood sample from the finger is drastically reduced. A photometric measurement principle is used to extract the glucose concentration from the blood sample, where a camera observes the chemical reaction between the chemical test strip and the blood sample.
Using methods of statistical signal processing and image processing, we are able to segment the produced images to detect the region where the chemical reaction takes place and estimate the intensity value of the region. This is then mapped to the underlying glucose concentration. Furthermore, we aim at reducing measurement time to increase the convenience of measurement for the patient. We use state-space approaches combined with a model of the chemical kinetics to achieve this.
For more information on this research project, contact Nevine Demitri.
Smartwatches or other wearable devices that contain optical heart rate monitors are an emerging technology that can be used, e.g as a tool to control the training load during physical exercises or to monitor physiologic conditions during daily activities. In contrast to the previous generation of devices, it is no longer necessary to wear an additional chest strap, since the heart rate is monitored from the wrist by means of photoplethysmography (PPG).
PPG refers to a noninvasive indirect measurement of the blood flow. Pulse oximeters illuminate the skin along with underlying blood vessels via light-emitting diodes (LEDs) to measure intensity changes of the reflected light that is absorbed by the photo diodes. Based on the intensity change in the PPG signal, the arterial oxygen saturation and the heart rate can be estimated. Combining the PPG signal with an electrocardiogram (ECG), allows the extraction of additional parameters, such as, e.g. the pulse arrival time (PAT) which is correlated with the blood pressure.
However, the PPG measurement is susceptible to motion artifacts, which inevitably occur during physical exercises. Motion induced artifacts can strongly deteriorate the quality of a PPG signal and signal processing techniques are sought for to remove the motion artifacts from the PPG signal prior to estimating the heart rate.
In this project, we develop new algorithms to robustly and accurately estimate the heart rate from PPG signals in the presence of motion artifacts. In order to apply our algorithms on real data, we also conduct measurement series and use external PPG data.