EEG-to-Report is a browser-based system designed to make routine electroencephalogram review produce structured training data for future clinical language models. Existing EEG tools generally focus on viewing and preprocessing signals, leaving hospitals without aligned waveform, feature and text examples for model development.
The framework accepts multiple EEG formats, standardizes channels and lets a reviewer annotate selected segments with typed notes or transcribed speech. For each segment, it calculates a consistent set of spectral, temporal, entropy, connectivity, spike-related and Hjorth features—a group of statistical measures commonly used to describe EEG signals. The result is stored with the clinical description in a portable JSON record.
An additional module combines an ensemble of convolutional neural networks with a language model to draft a narrative report that a neurologist can edit. The authors describe pilot annotations and a workflow foundation, not a clinical trial proving safe autonomous diagnosis. EEG interpretation depends on patient context, recording quality and expert judgment, and a polished draft can still contain a medically important error. The immediate contribution is data infrastructure: it connects a clinician’s review to reproducible feature-text pairs while keeping a person in the reporting loop. Larger, representative studies will be needed before the generated reports can support clinical decisions.