Generative deep learning for analysis of neural signals – a collaborative project between IBM Research and the NeuroTec center
Masterthesis / Biomedical Engineering / AI in Medicine
Status: Open
Time frame: not specified
Generative deep learning for analysis of neural signals – a collaborative project between IBM Research and the NeuroTec center
Project background
Epilepsy is a neurological disease which affects approximately 0.6%-0.8% of the world’s population. Innovative and promising treatment options, e.g. closed loop responsive neurostimulation, are continuously being developed to target patients suffering from pharmacoresistant epilepsy. Such emerging technologies often rely on precise real-time detection - and possibly forecasting - of seizures from biological signals, with iEEG being the most favorable candidate. We will be exploring a variety of architectures and alternatives to the common paradigm of deep learning in order to increase adoption by doctors and patients alike, to improve quality of life and therapeutic results. Such new techniques will need to exploit spatio-temporal relationship more efficiently and effectively than current approaches such as RNNs and extremely deep networks, especially in terms of power and sample efficiency.
Aim
Development of a novel iEEG and EEG deep-learning based algorithm, with a strong evaluation using real data collected at the Inselspital as the baseline.
Materials and Methods
This thesis consist of two parts.
The first part focuses on developing a novel algorithm to analyze and process iEEG and EEG data. This requires the student to 1) familiarize themselves with the relevant literature, 2) come up with novel approaches suited to the task. Great emphasis will be placed on brain-like computations, few-shot learning, and symbolic reasoning.
The second part of this project consists in producing a robust and extensive evaluation pipeline of all competitive approaches. The developed algorithm and corresponding findings are expected to be documented in form of a written thesis document. Outstanding work will be submitted at major conferences.
Figure 1 | Atlas amplifier from Neuralynx. | Source: https://neuralynx.fh-co.com
Nature of the Thesis
Development of algorithms: 50%
Evaluation: 50%
Requirements
Excellent programming skills in PyTorch
Good knowledge in time series analysis
Good knowledge of neural signals
Institutes
Department of Neurology, NeuroTec Center (www.neuro-tec.ch)
ARTORG Center (www.artorg.unibe.ch);
IBM Research Zurich (https://research.ibm.com/labs/zurich)
Supervisors
Dr. Francesco Carzaniga (IBM) & Dr. Abbas Rahimi (IBM)
Prof. Dr. Tobias Nef (Artorg)
Prof. Dr. med. Kaspar Schindler (NeuroTec)
Language: English
Contact: Prof. Dr. Kaspar Schindler, Rosenbühlgasse 25, CH-3010 Bern
Application: kaspar.schindler@insel.ch