Forecasting brain activity based on models of spatiotemporal brain dynamics: A comparison of graph neural network architectures

Wein, Simon; Schüller, Alina; Tomé, Ana Maria; Malloni, Wilhelm M.; Greenlee, Mark W.; Lang, Elmar

doi:10.1162/netn_a_00252

Cited by 10 publications

(9 citation statements)

References 126 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The Chebnet, a GCN, did appear to be better suited to model individual fMRI data. This result extends the similar conclusion of [11], [32] on BOLD auto-regression at the group level to the individual level. We hypothesize that the GCN's marginal improvement over linear models might be mainly attributable to better complexity efficiency and balance between spatial and temporal interactions on long time ranges, rather than the use of nonlinearities.…”

Section: Discussionsupporting

confidence: 88%

“…The superior performance of graph convolutional networks over VAR to model BOLD time-series is consistent with Wein's group studies [11], [32]. Our results extend these findings to the individual level.…”

Section: Model Comparison Key Features Of High Performance Modelssupporting

confidence: 90%

“…- -TFT: input_chunk_length [8,16,32] output_chunk_length [6] hidden_size [16,64] lstm_layers [1] num_attention_heads [4] full_attention [False] hidden_continuous_size [8] add_relative_index…”

Section: Iv: Dataset Versionsmentioning

confidence: 99%

See 2 more Smart Citations

A benchmark of individual auto-regressive models in a massive fMRI dataset

Paugam¹,

Pinsard²,

Lajoie³

et al. 2023

Preprint

View full text Add to dashboard Cite

Dense functional magnetic resonance imaging datasets open new avenues to create auto-regressive models of brain activity. Individual idiosyncrasies are obscured by group models, but can be captured by purely individual models given sufficient amounts of training data. In this study, we compared several deep and shallow individual models on the auto-regression of BOLD time series recorded during a natural video watching task. The best performing models were then analyzed in terms of their data requirements and scaling, subject specificity and the space-time structure of their predicted dynamics.We found the Chebnets, a type of graph convolutional neural network, to be best suited for BOLD auto-regression, closely followed by linear models. Chebnets demonstrated an increase in performance with increasing amounts of data, with no complete saturation at 9 h of training data. Significant subject specificity was found at short prediction time lags. The Chebnets were found to capture lower frequencies at longer prediction time lags, and the spatial correlations in predicted dynamics were found to match traditional functional connectivity networks.Overall, these results demonstrate that large individual fMRI datasets can be used to efficiently train purely individual auto-regressive models of brain activity, and that massive amounts of individual data are required to do so. The excellent performance of the Chebnets likely reflects their ability to combine spatial and temporal interactions on large time scales at a low complexity cost. Individual auto-regressive models have the potential to improve our understanding of the functional interactions in the human brain. This study is based on a massive, publicly-available dataset, which can serve for future benchmarks of individual auto-regressive modeling.

show abstract

Section: Discussionsupporting

confidence: 88%

Section: Model Comparison Key Features Of High Performance Modelssupporting

confidence: 90%

See 1 more Smart Citation

A benchmark of individual auto-regressive models in a massive fMRI dataset

Paugam¹,

Pinsard²,

Lajoie³

et al. 2023

Preprint

View full text Add to dashboard Cite

show abstract

“…As a result, a new field of geometric deep learning, graph neural networks (GNNs), has emerged, which has given the possibility to effectively process signals in the non-Euclidean geometry of graphs. Recently more and more GNNs have been proposed and applied in brain MRI analysis and disorder detection [21,22]. Parisot et al [23] introduced the spectral graph convolution network (GCN) with rfMRI and non-imaging data, representing populations as a sparse graph.…”

Section: Introductionmentioning

confidence: 99%

Early prediction of dementia using fMRI data with a graph convolutional network approach

Han,

Sun,

Zhao

et al. 2024

J. Neural Eng.

View full text Add to dashboard Cite

Objective: Alzheimer's disease is a progressive neurodegenerative dementia that poses a significant global health threat. It is imperative and essential to detect patients in the mild cognitive impairment (MCI) stage or even earlier, enabling effective interventions to prevent further deterioration of dementia. This study focuses on the early prediction of dementia utilizing Magnetic Resonance Imaging (MRI) data, using the proposed Graph Convolutional Networks (GCNs).Approach: Specifically, we developed a functional connectivity (FC) based GCN framework for binary classifications using resting-state fMRI data. We explored different types and processing methods of FC and evaluated the performance on the OASIS-3 dataset.We developed the GCN model for two different purposes: 1) MCI diagnosis: classifying MCI from normal controls; and 2) dementia risk prediction: classifying normal controls from subjects who have the potential for developing MCI but have not been clinically diagnosed as MCI.Main results: The results of the experiments revealed several important findings: First, the proposed GCN outperformed both the baseline GCN and Support Vector Machine (SVM). It achieved the best average accuracy of 80.3% (11.7% higher than the baseline GCN and 23.5% higher than SVM) and the highest accuracy of 91.2%. Secondly, the GCN framework with (absolute) individual FC performed slightly better than that with global FC generally. However, GCN using global graphs with appropriate connectivity can achieve equivalent or superior performance to individual graphs in some cases, which highlights the significance of suitable connectivity for achieving performance. Additionally, the results indicate that the self-network connectivity of specific brain network regions (such as default mode network, visual network, ventral attention network and somatomotor network) may play a more significant role in GCN classification.Significance: Overall, this study offers valuable insights into the application of GCNs in brain analysis and early diagnosis of dementia. This contributes significantly to the understanding of MCI and has substantial potential for clinical applications in early diagnosis and intervention for dementia and other neurodegenerative diseases. Our code for GCN implemention is available at: https://github.com/Shuning-Han/FC-based-GCN.

show abstract

“…However, to our knowledge, no previous studies have used a graph neural network (GNN), in conjunction with HD-EMG signals, to identify the movement/grasping that the amputee intends to perform. GNNs are helpful in a context where a high number of temporally-correlated spatial information is available [12]. This kind of neural network is composed of several propagation modules, which propagate information between nodes so that the aggregated information can capture both feature-based and topological information [13].…”

Section: Introductionmentioning

confidence: 99%

Graph Neural Networks for HD EMG-based Movement Intention Recognition: An Initial Investigation

Massa

Riboni

Nazarpour

2022

2022 IEEE International Conference on Recent Advances in Systems Science and Engineering (RASSE)

View full text Add to dashboard Cite

Recently, high-density (HD) EMG electrodes have been proposed for improving amputees' movement/grasping intention recognition, exploiting different machine learning techniques. HD EMG electrodes are composed of a large number of closely spaced channels that simultaneously acquire EMG signals from different parts of the muscle. Given the topological properties of these devices, it is important to fully exploit the spatiotemporal information provided by the electrodes to optimize recognition accuracy. In this work, we introduce the use of Graph Neural Networks (GNNs) to process HD EMG data for movement intention recognition of people with an amputation affecting the upper limbs and which use a robotic prosthesis. In this initial investigation of the approach, we conducted experiments using a real-world dataset consisting of EMG signals collected from 20 volunteers while performing 65 different gestures. We were able to detect 45 gestures with a classification error rate of less than 10%, and obtained an overall classification error rate of 8.75% with a standard deviation of 4.9. To the best of our knowledge, this is the first work in which GNNs are used for processing HD EMG data.

show abstract

Forecasting brain activity based on models of spatiotemporal brain dynamics: A comparison of graph neural network architectures

Cited by 10 publications

References 126 publications

A benchmark of individual auto-regressive models in a massive fMRI dataset

A benchmark of individual auto-regressive models in a massive fMRI dataset

Early prediction of dementia using fMRI data with a graph convolutional network approach

Graph Neural Networks for HD EMG-based Movement Intention Recognition: An Initial Investigation

Contact Info

Product

Resources

About