Influence of cognitive networks and task performance on fMRI-based state classification using DNN models.

Area of Science:

• Cognitive Neuroscience
• Machine Learning
• Neuroimaging

Background:

• Deep neural networks (DNNs) offer powerful data analysis but lack interpretability in cognitive neuroscience.
• Limited interpretability hinders the application of DNNs for understanding cognitive mechanisms from neuroimaging data.

Purpose of the Study:

• To apply interpretable DNNs for classifying cognitive task states from fMRI data.
• To investigate the cognitive underpinnings and individual variability in task performance using DNNs.

Main Methods:

• Employed a 1D convolutional neural network (1D-CNN) and a bidirectional long short-term memory (BiLSTM) network.
• Classified cognitive task states from fMRI data, analyzing feature importance and model performance.
• Correlated classification accuracy with individual cognitive performance metrics.

Main Results:

• 1D-CNN achieved 81% accuracy (Macro AUC=0.96), BiLSTM achieved 78% (Macro AUC=0.95).
• Both models showed significant correlation between prediction accuracy and individual cognitive performance (p<0.05 for 1D-CNN, p<0.001 for BiLSTM).
• Feature importance highlighted visual networks, with attention and control networks also showing high relevance. Lower accuracy correlated with poorer performance.

Conclusions:

• Interpretable DNNs can effectively classify cognitive states from fMRI data.
• Task-related neural activity, particularly in visual networks, is crucial for state classification.
• DNNs reveal relationships between neural activity, cognitive performance, and individual differences.