Dynamic networks differentiate the language ability of children with cochlear implants

Abstract

Background: Cochlear implantation (CI) in prelingually deafened children has been shown to be an effective intervention for developing language and reading skill. However, there is a substantial proportion of the children receiving CI who struggle with language and reading. The current study-one of the first to implement electrical source imaging in CI population was designed to identify the neural underpinnings in two groups of CI children with good and poor language and reading skill.

Methods: Data using high density electroencephalography (EEG) under a resting state condition was obtained from 75 children, 50 with CIs having good (HL) or poor language skills (LL) and 25 normal hearing (NH) children. We identified coherent sources using dynamic imaging of coherent sources (DICS) and their effective connectivity computing time-frequency causality estimation based on temporal partial directed coherence (TPDC) in the two CI groups compared to a cohort of age and gender matched NH children.

Findings: Sources with higher coherence amplitude were observed in three frequency bands (alpha, beta and gamma) for the CI groups when compared to normal hearing children. The two groups of CI children with good (HL) and poor (LL) language ability exhibited not only different cortical and subcortical source profiles but also distinct effective connectivity between them. Additionally, a support vector machine (SVM) algorithm using these sources and their connectivity patterns for each CI group across the three frequency bands was able to predict the language and reading scores with high accuracy.

Interpretation: Increased coherence in the CI groups suggest overall that the oscillatory activity in some brain areas become more strongly coupled compared to the NH group. Moreover, the different sources and their connectivity patterns and their association to language and reading skill in both groups, suggest a compensatory adaptation that either facilitated or impeded language and reading development. The neural differences in the two groups of CI children may reflect potential biomarkers for predicting outcome success in CI children.

Keywords: age of intervention; cochlear implant; electrical source imaging (ESI); electroencephalography (EEG); language and reading.

FIGURE 1

Demographics details along with CELF (clinical evaluation of language fundamentals) and RISE (reading inventory and scholastic evaluation) scores for all children. Here, HA indicates hearing aids (although none were used in this study), CI, cochlear implant; M, males and F indicates females, *indicates the significant difference of p < 0.05 for a T-Test.

FIGURE 2

Illustration of the methodological pipeline implemented in the study.

FIGURE 3

(A) Relative power for all frequency bands for high language (HL) and low language (LL) cochlear implanted children when compared to normal hearing (NH) cohort. The significant difference (p-values < 0.05) between the groups is indicated with an asterisk. (B) Cortical and subcortical sources of alpha band activity (C) beta band activity and (D) gamma band activity with significantly higher coherence in the HL and LL groups when compared to the NH group. Here, the coordinates below each source indicates the MNI coordinates of the location with the highest coherence source amplitude.

FIGURE 4

Time resolved partial directed coherence (TPDC)–based directional connectivity analysis showing the effective connectivity between the sources of alpha, beta and gamma band activity with the direction of information flow for the HL and LL groups. Here, the r-values indicate the correlations between the TPDC based connectivity to the timing of hearing restoration intervention [Age of first CI (in red), age of bilateral CI (in yellow), age of hearing aid fitting (in green)] as detailed in Table 2.

FIGURE 5

Support vector machine (SVM) based prediction accuracy using coherence and TPDC–based connectivity values for CELF (clinical evaluation of language fundamentals) and RISE (reading inventory and scholastic evaluation) sub-scores. Here, the SVM could predict all language and reading test scores however CELF and RISE-subtest–word recognition and decoding (WRDC) obtained the highest accuracy among them. Here: other subtests of RISE includes VOC, vocabulary; MORPH, morphology; SEN, sentence processing; EFFIC, efficiency of basic reading comprehension; RCOMP, reading comprehension.

FIGURE 6

Summarized view of all the sources and connectivity highlighting the difference between high language and low language CI groups.