Brain-Inspired Spatio-Temporal Associative Memories for Neuroimaging Data Classification: EEG and fMRI.

Area of Science:

• Neuroscience
• Artificial Intelligence
• Data Science

Background:

• Human decision-making relies on integrating information, forming spatio-temporal associations in the brain.
• Real-world decisions often involve incomplete information (limited variables or time).
• The brain functions as a spatio-temporal associative memory.

Purpose of the Study:

• To apply the Spatio-Temporal Associative Memory (STAM) framework to neuroimaging data (EEG and fMRI).
• To develop STAM models capable of classification using partial data after training on complete data.
• To evaluate the generalization accuracy of STAM models in both temporal and spatial domains.

Main Methods:

• Utilized the NeuCube brain-inspired spiking neural network framework.
• Developed STAM models for electroencephalography (EEG) and functional magnetic resonance imaging (fMRI) data.
• Trained classification models on complete spatio-temporal datasets and tested recall using partial time series and/or variables.

Main Results:

• Trained STAM models demonstrated accurate recall and classification performance using only partial EEG or fMRI data.
• Both temporal and spatial association and generalization accuracies were evaluated.
• The study confirmed that STAM can generalize from complete to partial data effectively.

Conclusions:

• The developed STAM-EEG and STAM-fMRI models show potential for classifying neuroimaging data using partial information.
• This pilot study opens avenues for applying STAM to other neuroimaging modalities like longitudinal MRI.
• Future research will focus on STAM's application in diagnosing and prognosing brain conditions and discovering biomarkers.