Prof. Dr.-Ing. Bert Arnrich


Below you will find a list of current research topics at our Connected Healthcare Chair. For further information on our projects please contact the responsibles persons or Prof. Dr. Bert. Arnrich.

OCD Early Warning System and Sensor-Supported Treatment Options

More than 2% of the world‘s population are diagnosed with obsessive-compulsive disorder (OCD) once in their lifetime [1]. Patients suffer from recurrent or persistent thoughts, images, impulses or actions and they have the desire to resist it [2]. These obsessions and compulsion take more than one hour a day. Typical behavior includes, among others, the repeatedly checking on things, e.g. if the door is locked or that the oven is off. If at all, OCD is often diagnosed very late. Responsible for this problem is lack of knowledge about the illness or shame of the patient [3]. With the multitude of new connected devices (smartwatch, smartphone, fitness tracker) and other Internet of Things (IoT) gadgets, compulsive habits could be identified early (Use Case 1). Consequently, a cognitive behavior therapy (CBT) could be started rapidly after identification, as the chances of recovery at an early stage of OCD are much better [4]. The most reliable CBT is the exposure and response prevention (ERP) where the patient is exposed to his fears without being allowed to perform any compulsive actions. The patient should realize that his anxiety ceases without any avoidance strategies. These subjective perceptions can be supported by sensor-based measurements of various physiological parameters (Use Case 2).

[1] W. K. Goodman, D. E. Grice, K. A. Lapidus and B. J. Coffey, "Obsessive-Compulsive Disorder," Psychiatric Clinics of North America, vol. 37, no. 3, pp. 257-267, 2014.

[2] A. T. Carr, "Compulsive neurosis: A review of the literature," Psychological Bulletin, vol. 81, no. 5, pp. 311-318, 1974.

[3] "Obsessive-Compulsive Disorder," National Institure of Mental Health , January 2016. [Online]. Available: www.nimh.nih.gov/health/topics/obsessive-compulsive-disorder-ocd/index.shtml. [Accessed 5 July 2019].

[4] E. H. a. S. P. Elisabetta Burchi, "From Treatment Response to Recovery: A Realistic Goal in OCD," Int J Neuropsychopharmacol, vol. 21, no. 11, pp. 1007-1013, 2018.

Recent Project: OCD Early Warning System

The two central components of an OCD Early Warning System are an Indoor Positioning System (IPS) and an Activity Recognition System. With the help of different technologies, including distance measurement to nearby anchor nodes, e.g. WiFi access points or Bluetooth beacons, it is possible to identify a person’s position. This localization is needed to recognize the very same activity. As an example, the checking of the front door 5 times in a row can be a compulsive act whereby the closing of 5 different doors in a house in a short time does not indicate a pathological behavior. Consequently, both systems, an IPS and an activity recognition system, are needed.

Furthermore, the OCD Early Warning System will very unobtrusive so that the user is not biased. Therefore, common connected devices, e.g. smartwatches, smartphones, other IoT gadgets, are used for the data collection.

INALO - Intelligent Alarm Optimization for ICU's

The monitoring of vital signs in the intensive care unit (ICU) has significantly improved patient safety by alerting ICU staff when a parameter deviates from the normal range. However, up to 99% of these alarms are false-positive, resulting in desensitization of staff to critical alarms and several deaths per year.

The goal of the INALO project is to develop patient-specific and user-centered software based on Artificial Intelligence (AI) for the intelligent optimization of alarms generated by patient monitors. Alarms from the existing patient monitoring system in ICU can be prioritized in a patient-specific manner and false-positive alarms can be filtered out with the INALO system. For the development, available clinical data from the patient monitoring system and the electronic health record (EHR) will be combined and the latest advances in machine learning will be applied.

The intelligent combination of different data streams and subsequent AI-based processing will generate added value in patient treatment. With this approach, INALO is facing one of the central challenges of digitalization in medicine.

Digital Phenotyping for and Beyond Clinical Trials

Still medical sampling has a snapshot approach lacking the dynamical behaviour of a person’s physiology. Sensor technologies are able to provide metrics by means of active (prompted) or passive (unnoticed) measurements, offering considerable flexibility in approach. Those high-frequency longitudinal data sets can then be used for prevention or characterization of a disease [1]. We address those time dependent features using monitoring of vital signs pre-, peri- and post- intervention. In various studies together with the department of integrative natural medicine (Charité) [2], German Institut of Nutrition (DIfE), Max Delbrück Centre for Biomedical Research (MDC, BIH) and Luxembourg Center for Systems Biomedicine we approach to better understand human phenotypes.  

[1] Kourtis LC, Regele OB, Wright JM, Jones GB. Digital biomarkers for Alzheimer’s disease: the mobile/wearable devices opportunity [In- ternet]. Vol. 2, npj Digital Medicine. 2019. Available from: http://dx.doi.org/https://doi. org/10.1038/s41746-019-0084-2

[2] Steckhan, Nico; Arnrich B. Quantified Complementary and Alternative Medicine : Convergence of Digital Health Technologies and Complementary and Alternative Medicine. Complement Med Res. 2020;8–10. 

Closed-loop Warning System for Epilepsy Monitoring

Epilepsy is the most common neurological disorder characterized by unprovoked and unexpected electrical bursts in the brain, which results in seizures. Initially, most epilepsy treatments lie on antiepileptic drugs (AEDs). However, among the 0.5-1% of the pediatric population suffering from epilepsy, about 25-30% of patients have drug-resistant epilepsy, which is defined as the failure of adequate trials of two tolerated, appropriately chosen and administered AEDs (whether as monotherapy or in combination) to achieve seizure freedom. The severity of the seizure can lead to SUDEP (sudden unexpected death in epilepsy). Additionally, the quality of life (QoL) of these patients highly depends on other comorbidities such as cognitive impairment, depression, medication side effects, sleep quality, and privacy, and importantly, the unpredictability of seizure occurrence. A closed-loop warning system is indispensable to improve the QoL of these patients, which includes continuous monitoring of patient's data, seizure prediction (before seizure onset) and detection (during seizure onset), maintaining electronic seizure diary, and providing right dosages of AEDs.

So far, video-Electroencephalography (EEG) is the gold standard for monitoring seizures, which is obtrusive and restricts the patient's daily life. Therefore, wearable technology is evolving in this field to provide an unobtrusive measurement. Within the framework of this project, pre-ictal and inter-ictal heart rate data will be collected from a wearable sensor to give a non-obtrusive and non-EEG based seizure prediction method in real-time. This system will provide an alarm to the patients or caregivers before seizure onset, which will eventually replace the patient-reported seizure diary with an automatic one. The AED dosages will be provided based on this diary, and the effects of AEDs on the seizure frequency will be analyzed.

Federated Learning

Machine learning has been widely adopted in a variety of fields as a means to gain information about data and make predictions. Open issues in a lot of areas of application are data privacy and lack of a sufficient amount of training data. In healthcare for example, patient data is heavily protected under the General Data Protection Regulation (GDPR), making artificial intelligence-driven research difficult. The newly proposed federated learning approach [1] has already shown promising results for privacy-preserving distributed machine learning systems. It allows a number of clients to jointly train a model on a central server without the need to transfer any sensitive information, all data stays on the clients‘ computers. Instead the model itself is distributed, periodically updated by each participant, and aggregated by the server.

In this project, I am developing new methods for building powerful predictive models for healthcare using private and distributed patient data. This data ranges from physiological signals collected in the hospital and electronic health records (EHR) to health data collected continuously in daily life by wearables and smartphones.

We are currently looking for a master student interested in writing his thesis in cooperation with Philips in Hamburg. If you want to learn more about this opportunity, please feel free to call or send an email.

This research is partially funded by the Federal Ministry of Education and Research of  Germany in the framework of KI-LAB-ITSE (project number 01IS19066).

[1] McMahan, H.B., Moore, E., Ramage, D., & Arcas, B.A. (2016). Federated Learning of Deep Networks using Model Averaging. ArXiv, abs/1602.05629.