Multimodal Digital Biomarkers for Early Brain Health Monitoring
Background:
Digital biomarkers derived from wearable, contactless, and smartphone-based technologies offer new opportunities for continuous and real-world monitoring of brain health. In the Swiss BrAInHealth project, we focus on sensing technologies and digital biomarkers, including smartphone interaction patterns, gait parameters, sleep and respiration measures, radar-based activity monitoring, and neuro-ophthalmological assessments. These digital biomarkers are intended to complement clinical evaluations and provide input to the Augmented Intelligence Platform.
Changes in mobility, sleep, activity patterns, and digital behavior may reflect early alterations in cognitive, motor, or affective function. However, individual digital biomarkers are often noisy and context-dependent. Combining multiple sensing modalities may improve the robustness and clinical usefulness of brain health monitoring. Developing data-driven methods to extract, harmonize, and analyze multimodal digital biomarkers is therefore an important step toward AI-supported dementia risk assessment and personalized prevention.
Aim
The aim of this thesis is to develop and evaluate a data analysis pipeline for multimodal digital biomarkers relevant to early brain health monitoring. The thesis will focus on extracting meaningful features from one or more sensing modalities and investigating how these features can be combined to characterize patterns associated with cognitive health, lifestyle, or dementia-related risk factors.
Description
The student will begin with a literature review on digital biomarkers for cognitive decline and brain health, focusing on sensing modalities such as gait analysis, sleep monitoring, smartphone-based digital phenotyping, contactless activity monitoring, and wearable sensor data.
The practical work will involve preprocessing and feature extraction from available sensor datasets or pilot data collected within the Swiss BrAInHealth context. Depending on data availability, the student may work with gait parameters, sleep metrics, activity patterns, smartphone interaction features, or a combination of these modalities. The student will implement methods for data cleaning, segmentation, feature extraction, and exploratory analysis. Machine learning or statistical models may then be used to identify patterns, cluster participants, detect anomalies, or predict clinically relevant outcomes.
The thesis may also investigate multimodal fusion strategies, such as early fusion of features, late fusion of model outputs, or interpretable dimensionality reduction methods. The resulting pipeline should be designed to generate structured digital biomarker outputs that could later be integrated into the Augmented Intelligence Platform.
Material and Methods:
- Data analysis and feature engineering: 40%
- Development of algorithms/pipeline: 40%
- Literature review: 20%
Requirements:
- Good programming skills in python
- Good knowledge of data analysis and machine learning
- Interest in wearable sensors, digital biomarkers, and biomedical signal/data analysis
- Basic knowledge of statistics and visualization
Institute:
ARTORG Center for Biomedical Engineering Research, University of Bern, Gerontechnology and Rehabilitation Group