We do not have any open positions at the moment.
Deep learning techniques and in particular convolutional neural networks (CNNs) are increasingly used in the field of neuroscience. CNNs have promising clinical applications in automating the detection of pathological patterns of neural activiy from large amounts of data.
One tool that is commonly used to measure neural functions is electroencephalography -EEG-. EEG is a non-invasive technique that measures time-series of electric activity of the brain, though electrodes placed on the scalp. EEG is used as a diagnostic tool in neurological disorders, like epilepsy or coma. EEG recordings in coma patients carry information about the integrity of neural functions in the absence of consciousness, and can be used to predict the patients’ outcome and chances of survival. However, there is a lack of automatic techniques for detecting pathological patterns of EEG responses to external stimuli during coma.
This master thesis will use CNNs for classifying EEG responses of coma patients to sounds, and for linking those to patients’ outcome. In our previous work, we have developped a CNN-based pipeline for classifying EEG resposnes of healthy volunteers. This master thesis will build upon this pipeline and will fine-tune its parameters and CNNs architecture for the specific case of EEG data of coma patients, whose EEG responses have higher variability than those of healthy participants.
The student working on this project will gain experience in deep learning algorithms for biomedical data, data augmentation techniques, and in the development of clinical biomarkers.
For more infomration please contact:
Florence Aellen: florence.aellen@inf.unibe.ch
Athina Tzovara: athina.tzovara@inf.unibe.ch
When we sleep, we process auditory stimuli from the environment around us. However, the neural mechanisms that allow the human brain to process stimuli during sleep are under-explored. We are collecting single-unit and local field potential (LFP) responses from patients with epilepsy, with intracranially implanted electrodes, with the aim of better characterizing how the auditory response is modulated by sleep.
This project, suitable for a master or bachelors thesis, will primarily focus on spike sorting of single-unit data and analysing firing rates and local field potentials in response to auditory stimuli.
The student working on it will gain experience with processing and analysis of a rich electrophysiological dataset and contribute to expanding the understanding of how information processing is altered by the absence of consciousness. This thesis will give experience with Python/matlab programming, processing and visualizing time-series signals.
For more infomration please contact:
Sigurd Alnes: sigurd.alnes@inf.unibe.ch
Athina Tzovara: athina.tzovara@inf.unibe.ch
When we sleep, we process auditory stimuli from the environment around us. However, the neural mechanisms that allow the human brain to process stimuli during sleep are under-explored. Recent studies suggest that responsiveness to external stimuli in sleep is related to complexity of neural activity and levels of neuronal noise.
Tools from the field of computer science, such as information theory or machine learning can be used to quantify the complexity of neural signals, recorded through electroencephalography (EEG). EEG is a non-invasive technique that measures electric activity of the brain through electrodes on the scalp. Measures of scale-free dynamics of neural activity can be applied on time-series of EEG signals to quantify the complexity of neural responses, or levels of neuronal noise. Interestingly, past studies suggest that the compexity of ongoing neural activity is modulated by sleep. However, it remains unknown how changes in neural complexity during sleep affect sensory processing of external stimuli (e.g. sounds).
The focus of this thesis will be on using techniques from information theory to study complexity of EEG responses to sounds during sleep and wakefulness. The student working on this project will analyse a dataset of high-density sleep EEG recordings using signal processing and information theory techniques. Also, the student will gain experience with Python programming, processing and visualizing time-series signals. No previous experience with sleep or EEG is necessary.
For more infomration please contact:
Sigurd Alnes: sigurd.alnes@inf.unibe.ch
Athina Tzovara: athina.tzovara@inf.unibe.ch
Electroencephalography (EEG) is a technique that measures electrical activity of the human brain, in invasive (iEEG) or non-invasive ways.
Classification algorithms have been applied in the field of neuroscience to analyse EEG responses to external stimuli. However, in the majority of cases, EEG responses are analysed for each participant separately. This allows us to examine how the brain responds to external stimuli of the environment, but without the possibility to study interactions with other humans. In a innovative recording setup, we have acquired iEEG data from patients with refractory epilepsy, who were performing simultaneously a non-verbal communication task. Our goal is to study the neural underpinnings of communication and transfer of knowledge from one patient to another.
This thesis will focus on analyzing an existing dataset of invasive electroencephalography recordings, using signal processing and machine learning techniques. The student working on this project will gain experience in advanced signal processing techniques for analyzing time-series data, invasive electroencephalography recordings and biomedical applications of classification algorithms. This project is in collaboration with Dr. Anais Llorens.
Please contact Athina Tzovara: athina.tzovara@inf.unibe.ch or Anais Llorens: anais@berkeley.edu