#MEG FILE EXTRACTOR TRIAL#

#MEG FILE EXTRACTOR TRIAL#

Typically, data are collected and pre-processed using a variety of techniques, such as cropping, trial averaging, normalization, band-pass filtering, and spatial transforms such as principle component analysis (PCA), independent component analysis (ICA) and common spatial patterns (CSP) 11, 12. Machine learning (ML) has been used in brain computer interfaces (BCIs), including for silent speech (imagined or covert) 8– 10.

We consider a dataset made up of a combination of a verb-generation task, the monosyllabic utterance / pah/ and the multisyllable non-word / pah tah kah/ with overlapping sets of subjects recorded using magnetoencephalography (MEG), to demonstrate age-related language and speech development. Although this process is less lateralized than higher level language function, non-word monosyllabic, and multisyllabic experiments still show some left-lateralization 7. Speech articulation divorced from language is much less lateralized in comparison to higher level language faculties regardless of age 1, and in terms of cortical systems, heavily engages the Rolandic cortex to integrate somatosensory and motor control representations to command the speech articulators 1.

A common experimental paradigm for demonstrating this is the verb-generation task, which requires a subject to produce a verb that is associated with an object/noun with which they have been provided (presented for example, using an image) 4. Children develop this lateralization over time, with young children typically showing bilateral representation of core language areas in young childhood, with increasing leftward lateralization in the pre-teen and teen years 4, 6. Language and speech in the adult brain spans a diverse set of regions and interconnections from brain stem to cerebral cortex 1– 3, but higher level language abilities are typically left-hemisphere lateralized for approximately 90% of the general population, although this can range from 70% to 95% in populations with distinct left and right handedness respectively 1, 4, 5. Speech development, from infancy to adulthood, is a remarkable and uniquely human process. Our proposed model scores over 95% mean cross-validation accuracy distinguishing above and below 10 years of age in single trials of un-seen subjects, and can classify publicly available EEG with state-of-the-art accuracy. Our data feature 92 subjects aged 4–18, recorded using a 151-channel MEG system. We highlight criteria the networks use, including relative weighting of channels and preferred spectro-temporal characteristics of re-weighted channels.

We present novel DNNs trained using raw magnetoencephalography (MEG) and electroencephalography (EEG) recordings that mimic the feature-engineering pipeline. Simple neural networks have been used extensively to classify data expressed as features, but require extensive feature engineering and pre-processing. Previous work has explored taking ‘deep’ neural networks (DNNs) designed for, or trained with, images to classify encephalographic recordings with some success, but this does little to acknowledge the structure of these data. Furthermore, we argue that the network makes predictions on the grounds of differences in speech development. We consider whether a deep neural network trained with raw MEG data can be used to predict the age of children performing a verb-generation task, a monosyllable speech-elicitation task, and a multi-syllabic speech-elicitation task.

Meg file extractor