R & T
Research
Doctoral Thesis
Statistical Modeling, Prototypical Implementation and Evaluation of a System for Fall-Risk Assessment of Elderly Persons
Falls are among the most frequent causes of injuries in the elderly and often have a severe impact on the phyiscal as well as mental state of persons concerned. Consequently, a lot of research has been done in order to find methods allowing for the early diagnosis of an increased fall-risk. To this aim, prospective study designs and standard methods of predictive statistics have been deployed to validate a broad variety of indicators suitable for the personal middle-term prediction of future falls. Although a satisfactory prediction accuracy has been achieved, these approaches suffer from major pragmatic drawbacks impeding their widespread application: On the one hand, the assessment is rather expensive in terms of required staff and/or technical equipment. On the other hand, patients find themselves in an "examination situation" , often percieved as being unpleasant.
In response to these issues, an alternative approach allowing for a permanent fall-risk monitoring using data of only one single accelerometer on board of a common wristwatch (originally featuring only an emergency call function) is developed. In this way, fall-risk assessment can be seamlessly integrated into persons' everyday lifes and requires only unexpensive sensor technology. Of course, these pragmatic achievements entail a higher complexity of the prediction models the algorithms automatically establishing prognoses are based on: Data acquisition takes place in a naturalistic (i.e. uncontrolled) setting and the measurement is of indirect nature, as no sensors are attached to the patients' lower extremities. After a supervised training to be performed one time for each patient, the multiple-stage procedure developed in the thesis reconstructs an already established fall-risk indicator, the so called gait variability: First, time intervals during which monitored persons were walking are determined from the permanently recorded motion data. For these intervals, gait events are reconstructed and gait variability is calculated, serving as input for the final computation of fall-risk. On the modelling level, methods from activtity classification and wavelet filtering are applied. As a major contribution beyond this, a theoretical analysis of the "best practice" in activity classification is provided.
Further Fields of Interest
- Predictive Statistics
- Formal Methods for the Design of Reactive Systems
(in particular: Model-based Testing) - Usability Testing
Teaching
- Introductory Cours in Stochastics
- Introduction to Software Usability
- Usability Testing
- Project Management
- Database Systems
- Tutorial in Database Systems
CV
since Oct. 2012
Researcher at the Chair of Information Science
2006-2012
Researcher at the AI-Chair of the University of Erlangen
- 2010-2012: Cooperation with SOPHIA AG ("Spitzencluster Medizintechnik")
- 2006-2010: Cooperation with AUDI AG
2000-2006
Studies in Computer Science at the University of Erlangen
Oct. 2004 - Febr. 2005
Internship at MEMODATA in Caen (France)
