Introducing the Joint EEG-Development Inference (JEDI) Model: A Multi-Way, Data Fusion Approach for Estimating Paediatric Developmental Scores via EEG.

Abstract

Accounting for developmental changes in children is a key consideration for adapting neurorehabilitation technologies to paediatric populations. Using well-established clinical tests and questionnaires can be resource and time intensive. With many data-driven rehabilitation approaches relying on EEG data, a means to rapidly assess and infer developmental status of children directly from these recordings could be critical. This paper proposes a new model for estimating classic developmental diagnostic scores by exploiting data fusion in a joint tensor-matrix decomposition of the EEG and score data. We have designated this model the joint EEG-development inference (JEDI) model. The proposed model is illustrated using a common EEG task (button press) via publicly available paediatric data from pre-adolescent children. Using three distinct recording blocks for training, validation, and testing and a ten-fold cross-validation scheme, a robust experimental design was used to evaluate the JEDI model under various conditions. Results indicate that the JEDI model can estimate the developmental scores of children while maintaining a high degree of similarity at a population level. These results highlight the JEDI model as a potential evolving tool for rapidly assessing child's development. Clinically, the proposed model could provide useful developmental information in a convenient and low resource manner.