Built on Rigorous Science
Every claim we make is backed by transparent methodology, conservative validation, and publicly available datasets. No shortcuts, no inflated metrics.
Validation Approach
LOSO Cross-Validation
Leave-one-subject-out cross-validation iterates through every subject in the dataset, training on all remaining subjects and testing on the held-out individual. This eliminates data leakage between train and test sets at the subject level, the most conservative validation strategy for diagnostic ML, simulating real clinical deployment where each patient is entirely new to the model.
Z-Score Normalization
Individual EEG spectral features are normalized against population-level baselines derived from the CHBMP normative dataset (282 healthy subjects). This transforms raw power values into standardised deviation scores, reducing the impact of inter-individual variability, recording conditions, and hardware differences on classification performance.
Why These Choices Matter
Much of the published EEG-ML literature reports inflated performance using k-fold cross-validation, which can leak within-subject information across folds. LOSO provides honest, clinically meaningful estimates. Z-score normalization against an independent normative dataset further ensures that models learn disorder-related patterns rather than site- or device-specific artifacts.
Performance Results
All results obtained using LOSO cross-validation. These are preliminary research metrics, not clinical performance claims.
Neuraxis-Child v1.0
Research PhaseAUC
~0.845
Balanced Accuracy
~0.802
Validation
LOSO CV
Dataset
IEEE ADHD
Age Range
5-17 years
Neuraxis-Adult v1.0
Research PhaseAUC
~0.727
Balanced Accuracy
-
Validation
LOSO CV
Dataset
TDBRAIN (N=72)
Age Range
18-35 years
Full validation report including sensitivity, specificity and MCC in preparation. All reported metrics are preliminary research results. Research Use Only (RUO).
Research Datasets
IEEE ADHD Dataset
Full dataset · Ages 5-17Public paediatric EEG dataset with clinical ADHD diagnoses. Chosen because it provides clinician-confirmed diagnostic labels (not self-report) and has been independently used in multiple published ADHD-EEG studies, enabling direct comparison with existing literature.
TDBRAIN
N=72 subset · Ages 18-35Adult EEG database from the University of Twente. Selected to test whether a model trained on paediatric data can generalise to adult populations recorded on different hardware. The 72-subject subset isolates the ADHD-relevant adult age range for focused validation.
CHBMP
282 subjects · Normative referenceCuban Human Brain Mapping Project. Large normative EEG dataset used to establish population-level spectral baselines. Enables Z-score normalization of individual patient features against a healthy reference distribution, improving cross-site and cross-device generalisation.
Publications
Manuscripts in preparation. Target journals include the Journal of Neural Engineering and IEEE Transactions on Neural Systems and Rehabilitation Engineering (TNSRE).
Commitment to Open Science
Synaption is committed to scientific reproducibility and transparency. All validation is performed on publicly available datasets. We report conservative LOSO metrics rather than optimistic k-fold estimates. Methodology details are documented to enable independent verification.