BrainBERT: Fine-tuning for Seizure Detection

We will show how BrainBERT, a foundation model trained on StereoEEG data, was fine‑tuned to detect seizures, covering data preparation, training pipeline, and performance evaluation.

Overview

This project was developed as a collaboration between the Brain Modulation Lab at MGH and InfoLab at MIT CSAIL. In this project, we fine tuned BrainBERT, a foundation model trained on StereoEEG data from epilepsy patients. In this demo, I will share how this model can be trained to detect seizures in open StereoEEG datasets.

Tech stack