CerebraGloss: Instruction-Tuning a Large Vision-Language Model for Fine-Grained Clinical EEG Interpretation

Links: OpenReview · Code

CerebraGloss studies how large vision-language models can support fine-grained clinical EEG interpretation. The work introduces an automated EEG-text data generation pipeline, an instruction-tuned LVLM, and CerebraGloss-Bench for open-ended EEG interpretation.

Highlights

• Builds a programmatic EEG-text instruction data engine with waveform, artifact, and background characterization.
• Trains a model for detailed waveform description, multi-choice reasoning, and multi-turn EEG dialogue.
• Evaluates the model on CerebraGloss-Bench and downstream clinical tasks such as seizure detection and sleep staging.