The Science Behind SynapCore
From raw EEG signals to clinically actionable insights. A transparent, validated, and reproducible diagnostic pipeline built on rigorous scientific methodology.
From Raw Signal to Clinical Insight
Five deterministic stages transform a multichannel EEG recording into a structured clinical decision support report.
Raw EEG Acquisition
Multi-channel EEG recordings captured using standardised clinical protocols. Support for 19-channel 10-20 montage and high-density configurations.
Preprocessing
Automated artifact rejection, bandpass filtering (0.5-45 Hz), ICA-based ocular artifact removal, bad channel interpolation, and re-referencing to common average.
Feature Extraction
Extraction of spectral power features (delta, theta, alpha, beta, gamma bands), coherence metrics, asymmetry indices, and connectivity measures across electrode pairs.
ML Classification
LightGBM gradient-boosted ensemble classifier with optimised hyperparameters. Trained and validated using leave-one-subject-out (LOSO) cross-validation for unbiased generalisation estimates.
Clinical Report
Structured clinical decision support report with probability scores, confidence intervals, feature importance visualisation, and normative comparison. Designed for clinician review, not autonomous diagnosis.
ML Architecture
SynapCore uses LightGBM, a gradient-boosted decision tree ensemble, selected for its interpretability, computational efficiency, and strong performance on structured tabular features extracted from EEG signals.
LightGBM Classifier
Gradient-boosted ensemble optimised for EEG spectral features. Provides feature importance rankings that map directly to neurophysiologically meaningful frequency bands, supporting clinical interpretability.
LOSO Cross-Validation
Leave-one-subject-out validation ensures that no data from the test subject appears during training. This provides the most conservative and clinically realistic estimate of generalisation performance, critical for diagnostic applications where overfitting must be eliminated.
Clinical Relevance
Unlike k-fold validation, LOSO simulates real-world deployment where the model encounters a previously unseen patient. This is why we report LOSO metrics rather than inflated k-fold scores common in EEG-ML literature.
Performance Metrics
Neuraxis validated on public research datasets using LOSO cross-validation. All metrics are preliminary research results.
| Metric | Neuraxis-Child v1.0Ages 5-17 | Neuraxis-Adult v1.0Ages 18-35 |
|---|---|---|
| AUC | 0.845 | 0.727 |
| Balanced Accuracy | 0.802 | - |
Full validation report including sensitivity, specificity and MCC in preparation. All reported metrics are preliminary research results. Research Use Only (RUO).
Research Datasets
Validated on publicly available, peer-reviewed datasets to ensure transparency and reproducibility.
IEEE ADHD Dataset
Publicly available paediatric EEG dataset used for ADHD classification research. Multi-channel recordings with clinical ADHD diagnoses. Used as primary training and validation set for Neuraxis-Child.
TDBRAIN
Adult EEG dataset from the Twente Database. Selected subset of subjects within the adult ADHD age range. Used for Neuraxis-Adult validation with LOSO cross-validation.
CHBMP
Cuban Human Brain Mapping Project. Large normative EEG reference dataset used for Z-score normalization and establishing population-level baselines for spectral features.
Regulatory Pathway
Synaption is committed to achieving full regulatory compliance before any clinical deployment. Current status: pre-commercial, Research Use Only.
IEC 62304
In progressSoftware lifecycle management for medical device software. Full traceability from requirements to validation.
CE Mark Class IIa
PlannedEuropean conformity assessment for medical devices with measuring function. Target: EU MDR 2017/745 compliance.
UKCA
PlannedUK Conformity Assessed marking for the UK market post-Brexit. Aligned with CE Mark Class IIa requirements.
FDA 510(k)
PlannedUS market clearance pathway. Predicate device identification and substantial equivalence demonstration in preparation.