BESA and University of Ansbach develop an improved AI-assisted method for automatic detection of epileptic seizures
From concept to fruition: the collaboration between BESA and the University of Ansbach turns ideas into tangible results. In their new joint project DEEP-EEG, which has now also received funding from the Bavarian Ministery of Economic Affairs, both partners develop together an improved AI-assisted method for automatic detection of epileptic seizures in standard EEG of adults.
In this interview, Fabienne Anselstetter, M.Sc., Researcher at BESA, and BESA project lead, Dr. Nicole Ille, explain how the collaboration came about and what is planned for the coming three years.
From left to right: Fabienne Anselstetter (BESA), Prof. Stefan Geißelsöder (Univ Ansbach), Dr. Nicole Ille (BESA), Prof. Christian Uhl (Univ Ansbach), Annika Stiehl (Univ Ansbach), Dr. Harald Bornfleth (BESA)
How did the project come about? Did BESA apply alone or was the application submitted jointly with the university?
“We have been working with Ansbach University of Applied Sciences for a long time – and we are very happy about this collaboration,” says Nicole. “The DEEP-EEG project for the detection of epileptic seizures using deep learning is a follow-up project of the previous successful cooperation with Prof. Christian Uhl and Prof. Stefan Geißelsöder during the 1st and 2nd Innovation Lab in 2021 and 2022. There we focused on the detection of interictal epileptic spikes and the explainability of deep learning.”
“That’s where I got in contact with BESA,” adds Fabienne who did her master thesis during the 1st Innovation Lab. “During the first Innovation Lab, we already worked with a variety of methods from the Machine Learning and Deep Learning area and data-engineering. Along with the comparison of the efficiency of the individual networks, we also focused on data-engineering. We examined the impact of differently prepared EEG input data regarding montage and filter-settings to the network’s performance. In addition, we further investigated the impact of different data reduction methods on the accuracy of the machine-learning networks.”
“That’s correct,” emphasizes Nicole. “For the follow-up DEEP-EEG project, BESA and Ansbach University applied together for a research grant in the Bavarian Collaborative Research Program of the Bavarian Ministery of Economic Affairs. We are very happy to receive this funding from January 2024 to January 2027.”
What criteria must be met in order to receive funding?
Nicole explains, “The funding of the Bavarian Collaborative Research Program is dedicated to industry-led, precompetitive joint research projects between small and medium-sized companies and research institutions cooperating in the area of information and communication technology. Company and research institution must be located in Bavaria.”
Using artificial intelligence to detect epileptic seizures sounds very advanced. What is the state of the art and how can your collaboration advance this field?
“Classical algorithms for seizure detection use heuristics based on temporal, spectral, statistical or non-linear EEG parameters. They can reach a sensitivity of 75-90% for the detection of epileptic seizures, however with a high false alarm rate of 0.1-5 per hour,” explains Nicole. “With modern deep-learning based approaches higher balanced accuracies can be reached. The major problem of seizure detection is the high variability of seizure patterns. Nonpathological EEG background rhythms and artifacts also complicate the detection. By combining the expertise of BESA in signal processing, artifact correction and classical seizure detection with the expertise of Ansbach University in machine learning and artificial intelligence we hope to further improve the accuracy of seizure detection for the benefit of patients with epilepsy.”
What is the ultimate goal of this project?
“The ultimate goal of this project is to develop an improved AI-assisted method for automatic detection of epileptic seizures in standard EEG of adults, which has a higher sensitivity than existing systems combined with at low false alarm rate. A low false alarm rate is especially important for the acceptance of patients and health-care providers. In a second step, we want to adapt the approach to just a few EEG electrodes in order to improve seizure detection in everyday life of seizure patients,” describes Nicole.
What are the particular challenges?
“The challenge of data-driven, deep-learning based approaches, in general, is the availability of a sufficient number of suitable data sets for the training of the neural network. In this project, we want to use open-source data as well as seizure EEG data from our own repository. We will also apply data augmentation techniques.”
Fabienne adds, “Modern neural networks can easily have millions of parameters. However, if the data set is too small, it is simply not enough to train the network sufficiently and results in overfitting. The model merely learns the data set through memorizing and therefore generalizes poorly, i.e. only delivers inadequate results for new data. Data augmentation is a process to increase the training data set by creating new but realistic data. For this, various techniques are used to generate artificial data points from existing ones.”
How will the collaboration between BESA and the University of Applied Sciences Ansbach work?
“Two thirds of the project tasks are executed by BESA. The distribution of the tasks is mostly according to the expertise of the project partners with a strong focus of BESA on spatio-temporal data processing and the development of an analysis pipeline for seizure detection. Both project partners will share the research work on suitable AI models for seizure detection. The best model will be included in the final analysis pipeline. As in our former collaboration projects, we are going to have regular project meetings, online and in presence, to share and discuss results,” says Nicole.
“I’m sure the project will run smoothly, based on our previous collaborations,” says Fabienne. “We will benefit from the synergy of the research capabilities of the university and the years of experience in EEG analysis and research at BESA. As in the previous Innovation Labs, we will work in close exchange and ensure the transfer of knowledge and expertise in terms of neuroscience and EEG.”
Are there milestones or a timetable?
“Yes, we have a project plan for 3 years with defined milestones,” describes Nicole. “At the end of every project year, we are going to report the progress of the project and the remaining tasks to the VDI Munich(1) as a project agency.”
A look into the future: how can this technology be used on the market in the future? What is your wish?
“After successful completion of the 3-year funded research phase the AI based prototype for seizure detection shall be integrated into a commercial software system and CE certified,” Nicole answers. “The software could be used in epilepsy centres to support medical professionals with the analysis of long-term EEG recordings. It could also be used directly by epileptic patients and caregivers outside the hospital for documentation and intervention of seizures.
Our wish is to develop a useful tool for the benefit of patients with epilepsy and their health-care team.”
“That’s true,” concludes Fabienne, “living with epilepsy is tough not knowing when and if another seizure will occur and can also be very restrictive in everyday life. After completing the project, I believe that we will be able to relieve the burden on medical staff as well as give patients and their relatives back a little more quality of life and safety through seizure detection.”
(1): VDI: Association of German Engineers
Comments are closed.