An Age-Adjusted EEG Source Classifier Accurately Detects School-Aged Barbadian Children That Had Protein Energy Malnutrition in the First Year of Life

Abstract

We have identified an electroencephalographic (EEG) based statistical classifier that correctly distinguishes children with histories of Protein Energy Malnutrition (PEM) in the first year of life from healthy controls with 0.82% accuracy (area under the ROC curve). Our previous study achieved similar accuracy but was based on scalp quantitative EEG features that precluded anatomical interpretation. We have now employed BC-VARETA, a novel high-resolution EEG source imaging method with minimal leakage and maximal sparseness, which allowed us to identify a classifier in the source space. The EEGs were recorded in 1978 in a sample of 108 children who were 5–11 years old and were participants in the 45+ year longitudinal Barbados Nutrition Study. The PEM cohort experienced moderate-severe PEM limited to the first year of life and were age, handedness and gender-matched with healthy classmates who served as controls. In the current study, we utilized a machine learning approach based on the elastic net to create a stable sparse classifier. Interestingly, the classifier was driven predominantly by nutrition group differences in alpha activity in the lingual gyrus. This structure is part of the pathway associated with generating alpha rhythms that increase with normal maturation. Our findings indicate that the PEM group showed a significant decrease in alpha activity, suggestive of a delay in brain development. Childhood malnutrition is still a serious worldwide public health problem and its consequences are particularly severe when present during early life. Deficits during this critical period are permanent and predict impaired cognitive and behavioral functioning later in life. Our EEG source classifier may provide a functionally interpretable diagnostic technology to study the effects of early childhood malnutrition on the brain, and may have far-reaching applicability in low resource settings.

Harmonized-Multinational qEEG Norms (HarMNqEEG)

“…Our work confirmed that a large part of the variance of traditional and HR DPs depends on frequency and age, underscoring the need for frequency and age-dependent norms to detect BDD. This fixed-effect dependency has been reported many times for traditional log-spectra, including several papers from our group (Taboada-Crispi et al, 2018; Bosch-Bayard et al, 2001; Bringas Vega et al, 2019). Indeed, the shape of the norms or “developmental surfaces” for traditional measures hold up with the larger multinational sample.…”

“…To check the validity of our new harmonized qEEG norms, revisited the problem of using the qEEG to classify school-children who suffered from Protein Energy Malnutrition limited to the first year of life (BMal) and to distinguish them from healthy classmate controls (BCtrl), who were matched by age, sex and handedness. Prior work is described in Bringas Vega et al (2019) Taboada-Crispi et al (2018). This work is part of the Barbados Nutrition Study-- a project that is still ongoing for nearly half a century (Galler et al, 1983a, 1983b).…”

“…We evaluate the discriminatory power of z-score types with the elastic-net regression using a stable sparse classifier (Bosch-Bayard et al,2018). The stable sparse classifier, also used for this same dataset in Bringas Vega et al (2019), uses multiple resampling to select a stable set of predictors and then (with additional independent resampling) estimate the ROC curves for discrimination between both groups. We plot the ROC curve and report the Area under the ROC curve (AUC) for each type of z-score.…”

“…Its first wave provided norms (means and standard deviations (SDs)) for the narrowband log-spectral DP based on 211 subjects from age 5 to 97. Despite being based on a single country database, CHBMP norms describe BDD consistently in other countries (Bosch-Bayard et al, 2001; Bringas Vega et al, 2019; Taboada-Crispi et al, 2018). Nevertheless, the relatively small sample sizes make country comparisons relatively underpowered compared to neuroimaging efforts such as ENIGMA (Thompson et al, 2014).…”

Resting State Healthy EEG: The First Wave of the Cuban Normative Database

“…• To calibrate and test new methods for source localization of activation and connectivity as in He et al (2019). • As the basis for the construction of classification models to predict abnormal EEG activity as in Bringas Vega et al (2019).…”