William Lehn-Schiøler og Magnus Guldberg Pedersen

Scaling Global Neurodiagnostics: Self-Supervised Learning for Point-of-Care EEG

Access to neurological diagnostics is severely limited in low-resource settings due to a shortage of trained neurologists. BrainCapture addresses this by providing affordable, smartphone-connected Electroencephalography (EEG) hardware, yet the challenge of expert data interpretation remains. This study explores Self-Supervised Learning (SSL) as a solution to bridge the diagnostic gap. By utilizing large-scale unlabeled brain activity to pretrain models, through contrastive learning, masked signal reconstruction, and denoising techniques, we can develop powerful feature extractors that require minimal expert labels for fine-tuning. These SSL methods enable BrainCapture’s platform to deliver high-accuracy automated screening for conditions like epilepsy, ensuring that sophisticated neuro-insight is accessible regardless of local clinical infrastructure.

Bio: William Lehn-Schiøler is an industrial PhD student at BrainCapture and DTU Health Tech working with Large EEG Models and AI Explainability. Magnus Guldberg Pedersen is a Master's student at BrainCapture and DTU Compute working with Large-Scale Biomedical Signal Databases and EEG modelling.