This toolkit provides age-based normalization and validation of EEG parameters across the human lifespan (ages 5-95). It implements linearized developmental norms derived from a large international cohort (N=1,563, resting state, eyes closed), enabling robust parameter estimation and brain age prediction.
This toolbox is the parametric version of the HarMNqEEG toolbox https://github.com/LMNonlinear/HarMNqEEG providing linearized norms for EEG periodic and aperiodic components, which enables rapid component-wise estimation of brain development and brain age with significantly fewer variables. The corresponding linearized norms are public in the ./data/ folder.
Analyzing a large international cohort (N=1,563, ages 5-95, resting state, eyes closed), we delineate two fundamental patterns:
- Aperiodic Activity: Monotonic decrease across the lifespan
- Periodic Activity: Growth-then-Decline trajectory
Both patterns show inflection points around age 20 and stabilize into senescence post-40.
Based on the logarithmic development progression, replotted parameter trajectories on a base-10 logarithmic age scale.
Aperiodic Parameters (Exponent, Offset):
- Near-linear Monotonic Decrease on log age scale
- Enables linear model approximation
Periodic Parameters (CF, PW):
- Approximately symmetric Growth-then-Decline path
- Centered on the inflection point
- Enables piecewise linear model approximation
- Model Simplification: Nonlinear developmental trajectories effectively approximated by linear/piecewise linear models
- Robust Estimation: Linearized functions provide more robust parameter estimates than nonlinear models without sacrificing accuracy
- Brain Age Prediction: Enables tractable brain age prediction by dividing into age groups
- Biomarker Potential: Linear trends of aperiodic parameters underscore their potential as reliable biomarkers of biological age
Linear Model:
Parameter = a 脳 log10(Age) + b
Where:
a: Slope (rate of change on log age scale)b: Intercept
Characteristics:
- Monotonic decrease across lifespan
- Stable, linear decline on log age scale
- Comparable to established molecular markers (telomere length, DNA methylation)
Piecewise Linear Model:
For Age < Inflection Point:
Parameter = a1 脳 log10(Age) + b1
For Age 鈮?Inflection Point:
Parameter = a2 脳 log10(Age) + b2
Where:
a1,b1: Growth phase parametersa2,b2: Decline phase parameters- Inflection point: ~25 years (CF) or ~30 years (PW)
Characteristics:
- Symmetric Growth-then-Decline pattern
- Piecewise linear on log age scale
- Captures development and aging phases
- Brain Age Prediction: Estimate biological age from EEG parameters
- Aging Assessment: Monitor aging trajectory and deviations
- Disease Detection: Identify abnormal aging patterns in neurological conditions
- Intervention Monitoring: Track response to therapeutic interventions
- Log Space Detection: Automatically detect if age data is in log space (max < 2)
- Space Conversion: Convert to linear space if needed using
power(10, x) - Model Fitting: Apply linear or piecewise linear models on log age scale
- Standardization: Calculate standardized residuals (z-scores)
- Semilog Plot: X-axis in log scale, Y-axis in linear scale
- Model Curve: Gray solid line showing model prediction
- Confidence Interval: Dashed lines (卤1 Std Dev) with shading overlay
- Data Points: Color-coded by disease/group with transparency
- Grid: Both major and minor gridlines for reference
- CSV Results: Predictions and standardized residuals
- PNG Plots: Semilog visualization with model curves
- Model Parameters: Fitted model coefficients and variance estimates
- Offset: Intercept of the power spectrum (reflects overall power level)
- Exponent: Slope of the power spectrum (reflects 1/f scaling)
- CF: Center Frequency (peak frequency of alpha oscillations)
- PW: Power (spectral power in alpha band)
- BW: Bandwidth (width of alpha peak)
from utility.params_age_log_validate import params_age_log_validate
# Run validation
results = params_age_log_validate(
model_path='./models',
result_dir='./result',
params=[
('Offset', None),
('Exponent', None),
('CF', 'alpha'),
('PW', 'alpha'),
('BW', 'alpha'),
],
testset_params_path_AP='./data/tdbrain_xi_para_mdd_adhd.csv',
testset_params_path_P='./data/tdbrain_alpha_band_para_mdd_adhd.csv'
)If you use this toolkit, please cite:
1. Li, M. et al. Aperiodic and Periodic EEG Component Lifespan Trajectories: Monotonic Decrease versus Growth-then-Decline. 2025.08.26.672407 Preprint at https://doi.org/10.1101/2025.08.26.672407 (2025).
2. Li, M. et al. Harmonized-Multinational qEEG norms (HarMNqEEG). NeuroImage 256, 119190 (2022).
---
## Data link:
1-The shared raw cross-spectra with encrypted ID is hosted at
Synapse.org (10.7303/syn26712693) and complete access is possible
after login in the system.
2-The multinational harmonized norms (HarMNqEEG norms) of traditional log-spectra and Riemannian cross-spectra
are hosted at Synapse.org (10.7303/syn26712979).
## Code link:
1-The corresponding HarMNqEEG code for calculating the z-scores based on
the HarMNqEEG norm opened in GitHub, see: https://github.com/LMNonlinear/HarMNqEEG.
2-The Robust Spectral parameterization toolkit is hosted at
GitHub (https://github.com/LMNonlinear/RobustSpectral).
---
**Version**: 1.0
**Last Updated**: 2025-10-06
**Status**: Production Ready
**Author**: Min Li, Ying Wang, Yaqi Chen
**opyright (c)**: 2020-2025 Min Li, minli.231314@gmail.com, Ying Wang, yingwangrigel@gmail.com,
**Hangzhou Dianzi Univerisity, Joint China-Cuba LAB, UESTC
**License: GNU General Public License v3.0 (see LICENSE file)