Print Close

Unraveling lifespan signatures: Time-series phenotyping and age prediction using MEG

Poster No:

1275

Submission Type:

Abstract Submission

Authors:

Christina Stier¹^,2, Elio Balestrieri¹^,2, Jana Fehring¹^,2, Andreas Wollbrink¹, Niels Focke³, Joachim Gross¹^,2

Institutions:

¹Institute for Biomagnetism and Biosignal Analysis, University of Muenster, Muenster, Germany, ²Otto Creutzfeldt Center for Cognitive and Behavioral Neuroscience, University of Muenster, Muenster, Germany, ³Clinic of Neurology, University Medical Center, Goettingen, Goettingen, Germany

First Author:

Christina Stier
Institute for Biomagnetism and Biosignal Analysis, University of Muenster|Otto Creutzfeldt Center for Cognitive and Behavioral Neuroscience, University of Muenster
Muenster, Germany|Muenster, Germany

Co-Author(s):

Elio Balestrieri
Institute for Biomagnetism and Biosignal Analysis, University of Muenster|Otto Creutzfeldt Center for Cognitive and Behavioral Neuroscience, University of Muenster
Muenster, Germany|Muenster, Germany

Jana Fehring
Institute for Biomagnetism and Biosignal Analysis, University of Muenster|Otto Creutzfeldt Center for Cognitive and Behavioral Neuroscience, University of Muenster
Muenster, Germany|Muenster, Germany

Andreas Wollbrink
Institute for Biomagnetism and Biosignal Analysis, University of Muenster
Muenster, Germany

Niels Focke
Clinic of Neurology, University Medical Center, Goettingen
Goettingen, Germany

Joachim Gross
Institute for Biomagnetism and Biosignal Analysis, University of Muenster|Otto Creutzfeldt Center for Cognitive and Behavioral Neuroscience, University of Muenster
Muenster, Germany|Muenster, Germany

E-Poster

Introduction:

Understanding the developing and aging brain is critical for predicting and assessing individual health risks. While the effects of age on the structural architecture of the brain have been studied previously for a wide age range (Bethlehem et al., 2022), similar efforts for neurophysiological dynamics have been sparse. In particular, it remains unclear which functional metrics predict individual age and can be used to estimate deviations from the normative lifespan.

Methods:

We extensively characterized 5 min of MEG resting-state signals derived from the Cambridge Center for Ageing and Neuroscience (CamCAN, Taylor et al., 2017) using 5988 features and whole-brain anatomical parcellation. We extracted features from the highly comparative time-series analysis toolbox (hctsa, Fulcher et al., 2017) and conventional features encompassing frequency-specific power, amplitude- and phase-based connectivity, the 1/f exponent (slope), and alpha peak frequency (instantaneous alpha peak frequency, center of gravity). Each feature was used to predict the individual age of participants (n = 350, 18-88 years) using partial least squares regression (PLS, 5 components) and 10-fold cross-validation (15 repetitions). The model performance was evaluated with the correlation between the true and predicted age (Pearson's r) and the mean absolute error (MAE), and potential confounding bias by sex and estimated intracranial volume (eTIV) tested (Spisak, 2022). To elaborate on the regional information of well-performing features, we applied k-means clustering on transformed PLS regression weights (Haufe et al., 2014).

Results:

Conventional features showed a prediction accuracy between r = 0.17 (MAE = 17.1) and r = 0.65 (MAE = 11.99), with the lowest performances of amplitude- and phase-coupling measures and best performances of measures depicting alpha peak frequencies (Figure 1). 107 features extracted from the hctsa toolbox outperformed conventional ones (r > 0.7), with the highest accuracies for the autocorrelation function at a time delay of 36 ms (lag 11; r = 0.75, MAE = 10.38, Figure 2). The prediction analyses for each feature were not confounded by sex or eTIV (p_FDR > 0.05). Clustering of the PLS weights for high-performing features (r > 0.73) showed that signal changes in the visual cortex, frontal-parietal areas, and central-temporal areas were most indicative of age.

·Figure 1

·Figure 2

Conclusions:

We provide an overview of the predictive value of neurophysiological metrics commonly studied to describe the healthy or diseased brain and a comprehensive set of time-series features in the context of development and aging. The profiling of regional MEG signals revealed lifespan patterns that can be captured by simple linear time-series measures, which were more accurate than the conventional spectral features. While further investigations on the relationship between informative features are needed to shed light on aging mechanisms, they all point to profound functional alterations in the visual, central-temporal, and frontal-parietal brain regions during adulthood.

Lifespan Development:

Aging

Lifespan Development Other ¹

Modeling and Analysis Methods:

Classification and Predictive Modeling

Novel Imaging Acquisition Methods:

MEG ²

Keywords:

Aging

MEG

Modeling

Other - Prediction

^1|2Indicates the priority used for review

Provide references using author date format

Bethlehem, R. A., Seidlitz, J., et al. (2022), Brain charts for the human lifespan. Nature, 604(7906), 525-533;

Fulcher, B. D. et al. (2017), hctsa: A computational framework for automated time-series phenotyping using massive feature extraction. Cell Systems, 5(5), 527-531;

Haufe, S., et al. (2014), On the interpretation of weight vectors of linear models in multivariate neuroimaging. Neuroimage, 87, 96-110.
Spisak, T. (2022), Statistical quantification of confounding bias in machine learning models. GigaScience, 11, giac082;

Taylor, J. R., et al. (2017), The Cambridge Centre for Ageing and Neuroscience (Cam-CAN) data repository: Structural and functional MRI, MEG, and cognitive data from a cross-sectional adult lifespan sample. Neuroimage, 144, 262-269