Analyzing pain perception is of high importance in numerous medical applications. For example, pain, or precisely the pain threshold, in physiotherapeutic treatments can not only determine the course or outcome of the treatment, but also shape the structure and composition of the exercises from the very beginning. Quantitative assessment of one's pain is traditionally based on self-assessment with questionnaires. However, for patients who are unable to communicate their pain (objectively), this method is not an option.

In collaboration with Prof. Kerstin Lüdtke (Physiotherapy, University of Lübeck), the Institute of Medical Informatics (IMI) is developing a learning-based pattern recognition platform that will automatically determine the current pain level using data from multiple wearable sensors. Data acquisition will be performed using multiple devices. Pain is induced by heat using a thermode, CHEPS (Contact Heat-Evoked Potential Stimulator), on the forearm of the study participants' non-dominant arm. During the experiment, participants are required to continuously report their perceived level of pain using a CoVAS (Computerised Visual Analogue Scale). To record the physical response of people in pain, two wearables (Empatica E4 and RespiBan) record a wide variety of physiological characteristics (such as BVP, EDA, EMG, ECG) during the test. Data from wearables, CoVAS and Thermode are synchronized and aligned with respect to sampling rate.

To classify pain vs. no pain (i.e., for a 2-class problem), a Random Forest procedure based on manually defined features is currently used. In addition, various Deep Learning approaches to automatic learning are being evaluated on the raw data. In the first pilot study with a dataset of 10 healthy subjects and a manually defined feature space (feature engineering), IMI achieved a classification rate of 76% on this 2-class problem. In the coming months, the Institute will significantly increase the volume of available training data, so that a significant improvement in the robustness of the classifier can be expected and additional pain classes can be added. An important aspect of research in this area is to develop such machine learning methods for solving the problem that provide new insights into the physiological and behavioral markers for pain.