Presentation
Search Abstracts | Symposia | Slide Sessions | Poster Sessions | Lightning Talks
Preoperative Neural Networks Predict Children’s Speech and Language Improvement 24 Months After Cochlear Implantation
There is a Poster PDF for this presentation, but you must be a current member or registered to attend SNL 2023 to view it. Please go to your Account Home page to register.
Poster E93 in Poster Session E, Thursday, October 26, 10:15 am - 12:00 pm CEST, Espace Vieux-Port
Di Yuan1, Nancy Young2,3, Patrick Wong1; 1The Chinese University of Hong Kong, 2Northwestern University, 3Ann and Robert H Lurie Children’s Hospital of Chicago
A previous study found that the preoperative neural features of 37 pediatric cochlear implant (CI) users predicted the variability of their speech perception development six months after surgery. Additionally, the neuroanatomical network unaffected by auditory deficit produced more precise predictions than the affected network. Given the experience-dependent plasticity of the human brain, the neuroanatomical network supporting speech and language development in children with CI may dynamically change across different linguistic developmental stages. It is hypothesized that children using CI for six months undergo an acclimatization process, involving higher-level brain regions (unaffected network) that invoke global attention to speech. Consequently, after this acclimatization period, children's speech and language abilities may extend from the auditory to the phonological stage, where the affected network also contributes to predicting long-term post-CI outcomes. This study aims to address this issue by utilizing preoperative neuroanatomical features to predict speech and language development in children using CI for 24 months. The study included forty-two children with sensorineural hearing loss who were CI candidates. The multivoxel pattern similarity of gray matter (GM) density was calculated in children with sensorineural hearing loss, in comparison with age-matched typical-hearing children. The GM similarity was used to predict speech and language improvement from pre-CI to 24 months post-CI. Separate machine-learning models were constructed for the whole-brain, affected, and unaffected brain networks. The nested k-fold cross-validation procedure and support vector classification (SVC) were employed. The classification accuracy, sensitivity, and specificity were calculated using a bootstrapping and permutation test. The results showed that speech and language improvement could be predicted by the whole brain GM similarity (accuracy: 69%; sensitivity: 68%; specificity: 67%). The unaffected network exhibited significantly higher accuracy than the affected network for 24-month improvement after surgery (affected: accuracy = 67%; unaffected: accuracy = 69%; p < 0.001). Furthermore, the accuracy of the model using non-neural features (household income, gender, prematurity) to predict postoperative improvement is 56%. Our results demonstrated that an extensive network including areas affected and unaffected by hearing loss contributes to the long-term speech and language development in pediatric CI users, despite slight but significant advantage of unaffected areas. This study suggests that there is a dynamic speech and language development supported by different preoperative brain networks in children with CI. Further studies are needed to delve into the relationship between neuroanatomical networks and the various linguistic components in children with CI.
Topic Areas: Language Development/Acquisition, Computational Approaches