MEGnet: Automatic ICA-based artifact removal for MEG using spatiotemporal convolutional neural networks

Treacher, Alex; Garg, Prabhat; Davenport, Elizabeth M.; Godwin, Ryan C.; Proskovec, Amy L.; Bezerra, Leonardo Guimaraes; Murugesan, Gowtham Krishnan; Wagner, Benjamin; Whitlow, Christopher T.; Stitzel, Joel D.; Maldjian, Joseph A.; Montillo, Albert

doi:10.1016/j.neuroimage.2021.118402

Cited by 14 publications

(5 citation statements)

References 39 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Heuristics such as the correlation of components to the EOG/ECG signals can help with this, but are not always present. Helpfully, methods which use machine learning methods to identify the origin of a component are starting to become available for MEG data (Treacher et al, 2021). Finally, a third potential issue is the assumption the spatial topographies of these artefacts are similar across all frequencies.…”

Section: Discussionmentioning

confidence: 99%

Is high-frequency activity evidence of an anterior temporal lobe network or micro-saccades?

O'Neill

Mellor

Seymour

et al. 2023

Preprint

View full text Add to dashboard Cite

There is renewed interest in electrical activity that extends beyond the typical electrophysiological 100 Hz bandwidth. This activity, often in the anterior temporal lobe, has been attributed to processes ranging from memory consolidation to epileptiform activity. Here, using an open-access resting state magnetoencephalography (MEG) dataset (n = 89), and a second task-based MEG dataset, we could reliably localise high-frequency power to the temporal lobes across multiple bands up to 300-400 Hz. A functional connectivity analysis of this activity revealed a robust resting state bilateral network between the temporal lobes. However, we also found robust coherence in the 100-200 and 200-300 Hz bands between source reconstructed MEG data and the electrooculography (EOG) localised to within the temporal poles. Additional denoising schemes applied to the data could reduce power localisation to the temporal poles but the topography of the functional network did not drastically alter. Whilst it is clear that this network is biological and robust to established denoising methods, we cannot definitively rule yet on whether this is of neural or myogenic origin.

show abstract

Section: Discussionmentioning

confidence: 99%

Is high-frequency activity evidence of an anterior temporal lobe network or micro-saccades?

O'Neill

Mellor

Seymour

et al. 2023

Preprint

View full text Add to dashboard Cite

show abstract

“…The choice of architecture depends on the task that the model is intended to perform. As for the hyperparameters, automated selection procedures have been developed but are rarely used in neuroimaging applications (e.g., Treacher et al, 2021 ).…”

Section: Discussionmentioning

confidence: 99%

Deep learning in neuroimaging data analysis: Applications, challenges, and solutions

Avberšek

Repovš

2022

Front. Neuroimaging

View full text Add to dashboard Cite

Methods for the analysis of neuroimaging data have advanced significantly since the beginning of neuroscience as a scientific discipline. Today, sophisticated statistical procedures allow us to examine complex multivariate patterns, however most of them are still constrained by assuming inherent linearity of neural processes. Here, we discuss a group of machine learning methods, called deep learning, which have drawn much attention in and outside the field of neuroscience in recent years and hold the potential to surpass the mentioned limitations. Firstly, we describe and explain the essential concepts in deep learning: the structure and the computational operations that allow deep models to learn. After that, we move to the most common applications of deep learning in neuroimaging data analysis: prediction of outcome, interpretation of internal representations, generation of synthetic data and segmentation. In the next section we present issues that deep learning poses, which concerns multidimensionality and multimodality of data, overfitting and computational cost, and propose possible solutions. Lastly, we discuss the current reach of DL usage in all the common applications in neuroimaging data analysis, where we consider the promise of multimodality, capability of processing raw data, and advanced visualization strategies. We identify research gaps, such as focusing on a limited number of criterion variables and the lack of a well-defined strategy for choosing architecture and hyperparameters. Furthermore, we talk about the possibility of conducting research with constructs that have been ignored so far or/and moving toward frameworks, such as RDoC, the potential of transfer learning and generation of synthetic data.

show abstract

“…The data were then downsampled to 256 Hz and decomposed into 30 components using InfoMax Independent Component Analysis (ICA) 28 . The number of components for the ICA decomposition was chosen based on previous research 29 , 30 . Eyeblink and cardiac artifacts were visually identified based on their topographic map and time series, then subsequently rejected (mean number of rejected components = 1.94).…”

Section: Methodsmentioning

confidence: 99%

Age-related increases in right hemisphere support for prosodic processing in children

Li,

Dimitrijevic,

Gordon

et al. 2023

Sci Rep

View full text Add to dashboard Cite

Language comprehension is a complex process involving an extensive brain network. Brain regions responsible for prosodic processing have been studied in adults; however, much less is known about the neural bases of prosodic processing in children. Using magnetoencephalography (MEG), we mapped regions supporting speech envelope tracking (a marker of prosodic processing) in 80 typically developing children, ages 4–18 years, completing a stories listening paradigm. Neuromagnetic signals coherent with the speech envelope were localized using dynamic imaging of coherent sources (DICS). Across the group, we observed coherence in bilateral perisylvian cortex. We observed age-related increases in coherence to the speech envelope in the right superior temporal gyrus (r = 0.31, df = 78, p = 0.0047) and primary auditory cortex (r = 0.27, df = 78, p = 0.016); age-related decreases in coherence to the speech envelope were observed in the left superior temporal gyrus (r = − 0.25, df = 78, p = 0.026). This pattern may indicate a refinement of the networks responsible for prosodic processing during development, where language areas in the right hemisphere become increasingly specialized for prosodic processing. Altogether, these results reveal a distinct neurodevelopmental trajectory for the processing of prosodic cues, highlighting the presence of supportive language functions in the right hemisphere. Findings from this dataset of typically developing children may serve as a potential reference timeline for assessing children with neurodevelopmental hearing and speech disorders.

show abstract

MEGnet: Automatic ICA-based artifact removal for MEG using spatiotemporal convolutional neural networks

Cited by 14 publications

References 39 publications

Is high-frequency activity evidence of an anterior temporal lobe network or micro-saccades?

Is high-frequency activity evidence of an anterior temporal lobe network or micro-saccades?

Deep learning in neuroimaging data analysis: Applications, challenges, and solutions

Age-related increases in right hemisphere support for prosodic processing in children

Contact Info

Product

Resources

About