SIFNet: Electromagnetic Source Imaging Framework Using Deep Neural Networks

Sun, Rui; Sohrabpour, Abbas; Ye, Shuai; He, Bin

doi:10.1101/2020.05.11.089185

Cited by 6 publications

(8 citation statements)

References 51 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…While various studies showed the strength and capability of such models to capture realistic brain dynamics, no prior work has employed these models to develop a noninvasive imaging and modeling framework and rigorously validated the results using ground truth, such as intracranial recordings and surgical resection outcomes in epilepsy patients. Our proposed deep learning framework ( 59 ) represents an effort at integrating biophysically inspired mesoscale models of neuronal activation, as embodied by NMMs, into DL-based ESI for distributed source imaging. The use of biophysically inspired brain network models for generating big training data represents an important undertaking as the performance, usability, and robustness of a neural network are based upon and bounded by the quality and nature of its training examples.…”

Section: Discussionmentioning

confidence: 99%

Deep neural networks constrained by neural mass models improve electrophysiological source imaging of spatiotemporal brain dynamics

Sun

Sohrabpour

Worrell

et al. 2022

Proc. Natl. Acad. Sci. U.S.A.

Self Cite

View full text Add to dashboard Cite

Many efforts have been made to image the spatiotemporal electrical activity of the brain with the purpose of mapping its function and dysfunction as well as aiding the management of brain disorders. Here, we propose a non-conventional deep learning–based source imaging framework (DeepSIF) that provides robust and precise spatiotemporal estimates of underlying brain dynamics from noninvasive high-density electroencephalography (EEG) recordings. DeepSIF employs synthetic training data generated by biophysical models capable of modeling mesoscale brain dynamics. The rich characteristics of underlying brain sources are embedded in the realistic training data and implicitly learned by DeepSIF networks, avoiding complications associated with explicitly formulating and tuning priors in an optimization problem, as often is the case in conventional source imaging approaches. The performance of DeepSIF is evaluated by 1) a series of numerical experiments, 2) imaging sensory and cognitive brain responses in a total of 20 healthy subjects from three public datasets, and 3) rigorously validating DeepSIF’s capability in identifying epileptogenic regions in a cohort of 20 drug-resistant epilepsy patients by comparing DeepSIF results with invasive measurements and surgical resection outcomes. DeepSIF demonstrates robust and excellent performance, producing results that are concordant with common neuroscience knowledge about sensory and cognitive information processing as well as clinical findings about the location and extent of the epileptogenic tissue and outperforming conventional source imaging methods. The DeepSIF method, as a data-driven imaging framework, enables efficient and effective high-resolution functional imaging of spatiotemporal brain dynamics, suggesting its wide applicability and value to neuroscience research and clinical applications.

show abstract

Section: Discussionmentioning

confidence: 99%

Deep neural networks constrained by neural mass models improve electrophysiological source imaging of spatiotemporal brain dynamics

Sun

Sohrabpour

Worrell

et al. 2022

Proc. Natl. Acad. Sci. U.S.A.

Self Cite

View full text Add to dashboard Cite

show abstract

“…finding the neural sources given the signals measured on the scalp, is of major interest both for basic research and in the clinical context (e.g., localization of seizure onset zones in epilepsy). ANN-based analysis methods are gaining attention in the past years and appear a viable option to solve the inverse problem of the M/EEG (Cui et al, 2019; Dinh, Samuelsson, Hunold, & Hämäläinen, 2019; Hecker et al, 2020; Huang et al, 2020; Pantazis & Adler, 2021; Sun et al, 2020; Tankelevich, 2019).…”

Section: Discussionmentioning

confidence: 99%

Evaluation of Long-Short Term Memory Networks for M/EEG Source Imaging with Simulated and Real EEG Data

Hecker

Rupprecht

Elst

et al. 2022

Preprint

View full text Add to dashboard Cite

Magneto- and electroencephalography (M/EEG) are widespread techniques to measure neural activity in vivo at a high temporal resolution but relatively low spatial resolution. Locating the sources underlying the M/EEG poses an inverse problem, which is itself ill-posed. In recent years, a new class of source imaging methods was developed based on artificial neural networks. We present a long-short term memory (LSTM) network to solve the M/EEG inverse problem. It integrates several aspects essential for qualitative inverse solutions: Low computational cost, exploitation of both spatial and temporal information, input flexibility and robustness to noise. Using simulation data, the LSTM shows higher accuracy on multiple metrics and for varying numbers of neural sources, compared to classical inverse solutions but also compared to our alternative architectures without integration of temporal information. It successfully integrates the temporal context given by sequential data, which is particularly useful with noisy data. Real data of a median nerve stimulation paradigm was used to show that the LSTM predicts plausible sources that are in concordance with classical inverse solutions. The performance of the LSTM regarding its robustness to noise renders it a promising and easy-to-apply inverse solution to be considered in future source localization studies and for clinical applications.

show abstract

“…Thus, the TBFs can not only increase the generalizability of the neural network but also reduce the complexity of the network training. SIFNet [28] also utilized a deep learning approach with data generation for the ESI problem, but there are three major differences between SIFNet and DST-DAE: 1) SIFNet only focuses on the single source estimation, while DST-DAE can estimate a multisource case; 2) SIFNet focuses on the source localization by reformulating the ESI problem as a classification problem, while DST-DAE further considers the regression of the temporal dynamics of the estimated sources; 3) SIFNet utilizes training signals corrupted with different SNRs to guarantee a modest denoising ability, leading to more sensitivity to the noise levels of the test samples, while DST-DAE has a superior and efficient denoising ability against different noise levels due to its denoising architecture.…”

Section: Discussionmentioning

confidence: 99%

“…Thus, deep neutal networks could be a potential new solution for the ESI problem. Recently, the source imaging framework network (SIFNet) [28] was proposed in an attempt to transform the ESI inverse problem into a supervised multiclassification task. After the brain source space is modeled as different interconnected regions, a residual-blockbased classification network can be succesfully trained using the generated training set, in which each training label is a one-hot vector representing the activation state of each region.…”

Section: Introductionmentioning

confidence: 99%

Electromagnetic Source Imaging via a Data-Synthesis-Based Denoising Autoencoder

Huang¹,

Liang²,

Liu³

et al. 2020

Preprint

View full text Add to dashboard Cite

Electromagnetic source imaging (ESI) is a highly ill-posed inverse problem. To find a unique solution, traditional ESI methods impose a variety of priors which may not reflect the actual source properties. Such limitations of traditional ESI methods hinder their further applications. Inspired by deep learning approaches, a novel data-synthesized spatio-temporal denoising autoencoder method (DST-DAE) method was proposed to solve the ESI inverse problem. Unlike the traditional methods, we utilize a neural network to directly seek generalized mapping from the measured E/MEG signals to the cortical sources. A novel data synthesis strategy is employed by introducing the prior information of sources to the generated large-scale samples using the forward model of ESI. All the generated data are used to drive the neural network to automatically learn inverse mapping. To achieve better estimation performance, a denoising autoencoder (DAE) architecture with spatio-temporal feature extraction blocks is designed. Compared with the traditional methods, we show (1) that the novel deep learning approach provides an effective and easy-to-apply way to solve the ESI problem, that (2) compared to traditional methods, DST-DAE with the data synthesis strategy can better consider the characteristics of real sources than the mathematical formulation of prior assumptions, and that (3) the specifically designed architecture of DAE can not only provide a better estimation of source signals but also be robust to noise pollution. Extensive numerical experiments show that the proposed method is superior to the traditional knowledge-driven ESI methods.

show abstract

SIFNet: Electromagnetic Source Imaging Framework Using Deep Neural Networks

Cited by 6 publications

References 51 publications

Deep neural networks constrained by neural mass models improve electrophysiological source imaging of spatiotemporal brain dynamics

Deep neural networks constrained by neural mass models improve electrophysiological source imaging of spatiotemporal brain dynamics

Evaluation of Long-Short Term Memory Networks for M/EEG Source Imaging with Simulated and Real EEG Data

Electromagnetic Source Imaging via a Data-Synthesis-Based Denoising Autoencoder

Contact Info

Product

Resources

About