Machine learning approaches for electroencephalography and magnetoencephalography analyses in autism spectrum disorder: A systematic review.

There are growing application of machine learning models to study the intricacies of non-linear and non-stationary characteristics of electroencephalography (EEG) and magnetoencephalography (MEG) data in neurobiologically complex and heterogeneous conditions such as autism spectrum disorder (ASD). Such tools have potential diagnostic applications, and given the highly heterogeneous presentation of ASD, might prove fruitful in early detection and therefore could facilitate very early intervention. We conducted a systematic review (PROSPERO ID#CRD42021257438) by searching PubMed, EMBASE, and PsychINFO for machine learning approaches for EEG and MEG analyses in ASD. Thirty-nine studies were identified, of which the majority (18) used support vector machines for classification; other successful methods included deep learning. Thirty-seven studies were found to employ EEG and two were found to employ MEG. This systematic review indicate that machine learning methods can be used to classify ASD, predict ASD diagnosis in high-risk infants as early as 3 months of age, predict ASD symptom severity, and classify states of cognition in ASD with high accuracy. Replication studies testing validity, reproducibility and generalizability in tandem with randomized controlled trials in ASD populations will likely benefit the field.

Cognitive control, interference inhibition, and ordering of information during working memory in younger and older healthy adults.

Investigating effects of aging on neurophysiological mechanisms underlying working memory provides a better understanding of potential targets for brain intervention to prevent cognitive decline. Theta-gamma coupling (TGC) indexes the ability to order information processed during working memory tasks. Frontal theta event-related synchronization (ERS) and parietal alpha event-related desynchronization (ERD) index cognitive control and interference inhibition, respectively. Relative contributions of TGC, theta ERS, and alpha ERD in relation to stimulus presentation are not characterized. Further, differential effect of normal aging on pre- or post-stimulus processes is unknown. Electroencephalography was recorded in 66 younger and 41 older healthy participants while performing 3-back working memory task. We assessed relationships between 3-back task performance and each of post-stimulus TGC, pre-stimulus parietal alpha ERD, and pre-stimulus frontal theta ERS in each age group. While older adults performed worse on 3-back task than younger adults, TGC, alpha ERD, or theta ERS did not differ between the two groups. TGC was positively associated with 3-back performance in both age groups; pre-stimulus alpha ERD was associated with performance among younger adults; and pre-stimulus theta ERS was not associated with performance in either group. Our findings suggest that both pre-stimulus interference inhibition and post-stimulus ordering of information are important for working memory in younger adults. In contrast, performance in older adults appears to depend only on post-stimulus ordering of information. These specific contributions of neurophysiological resources may explain the poorer performance of older adults and suggest different targets to enhance working memory in age groups.

Assessing the Longitudinal Relationship between Theta-Gamma Coupling and Working Memory Performance in Older Adults.

Theta-gamma coupling (TGC) is a neurophysiologic mechanism that supports working memory (WM). TGC is associated with N-back performance, a WM task. Similar to TGC, theta and alpha event-related synchronization (ERS) and desynchronization (ERD) are also associated with WM. Few studies have examined the longitudinal relationship between WM performance and TGC, ERS, or ERD. This study aimed to determine if changes in WM performance are associated with changes in TGC (primary aim), as well as theta and alpha ERS or ERD over 6 to 12 weeks. Participants included 62 individuals aged 60 and older with no neuropsychiatric conditions or with remitted Major Depressive Disorder (MDD) and no cognitive disorders. TGC, ERS, and ERD were assessed using electroencephalography (EEG) during the N-back task (3-back condition). There was an association between changes in 3-back performance and changes in TGC, alpha ERD and ERS, and theta ERS in the control group. In contrast, there was only a significant association between changes in 3-back performance and changes in TGC in the subgroup with remitted MDD. Our results suggest that the relationship between WM performance and TGC is stable over time, while this is not the case for changes in theta and alpha ERS and ERD.