·
11 commits
to main
since this release
Release v2.0.0 - SeizureTransformer Architecture & Repository Transition
🚀 Major Announcement
We are transitioning our active development to specialized repositories for better modularity:
- Clinical Evaluation & NEDC Scoring: Active development has moved to SeizureTransformer
- Core EEG Processing: Remains in this repository
What's New in v2.0.0
✅ SeizureTransformer Implementation
- Wu 2025 Architecture: Full CNN+Transformer implementation (
seizure_transformer_wu2025.py) - 41M Parameters: Production-ready model matching paper specifications
- TUSZ Support: Complete dataset pipeline for temporal seizure detection
- CI/CD Compliant: All type checking and linting issues resolved
📊 TUEV Event Classification Status
- Implementation: ✅ Verified correct, matches EEGPT reference exactly
- Performance: 22% BAC (vs 62% paper claim)
- Status: ARCHIVED - Dataset has fatal class imbalance (22 samples for minority class)
- Recommendation: Use TUAB or other balanced datasets
🔄 Repository Status
This repository (brain-go-brrr) will continue to maintain:
- EEGPT foundation model integration
- YASA sleep analysis (87% accuracy)
- Autoreject QC pipeline
- Core API infrastructure
For active development on seizure detection and clinical evaluation:
- Please visit: github.com/Clarity-Digital-Twin/SeizureTransformer
- Features there: NEDC scoring, clinical metrics, operating point optimization
Breaking Changes
- Removed
seizure_transformer_toy_deprecated.py(duplicate classes) - SeizureTransformer wrapper now uses Wu 2025 architecture by default
Migration Guide
If you're working on seizure detection:
# Clone the new repository
git clone https://github.com/Clarity-Digital-Twin/SeizureTransformer
cd SeizureTransformer
uv pip install -e .If you're using core EEG features:
# Continue using this repository
uv pip install brain-go-brrr==2.0.0Contributors
Thanks to the team for the SeizureTransformer implementation and architecture improvements.
Full Changelog
- Wu 2025 SeizureTransformer architecture implementation
- Complete TUSZ dataset support
- TUEV implementation verified but archived (poor performance)
- All mypy type checking issues resolved
- CI/CD pipeline fully passing
Full Changelog: v1.2.0...v2.0.0