Task-State fMRI-Derived Whole-Brain Functional Topology-Constrained Spiking Neural Network with an Embedded Auditory...

Area of Science:

• Computational Neuroscience
• Neuroimaging
• Artificial Intelligence

Background:

• Spiking neural networks (SNNs) are crucial for speech recognition, but their connectivity patterns often lack biological realism.
• Current SNN designs use random or manual connections, failing to capture the brain's functional organization during speech perception.

Purpose of the Study:

• To develop a novel SNN framework for speech recognition constrained by task-based functional Magnetic Resonance Imaging (fMRI) data.
• To investigate if a biologically inspired topology derived from fMRI enhances SNN performance and interpretability in speech recognition.

Main Methods:

• Utilized whole-brain functional topology from human fMRI data during audiobook listening as a prior for SNN recurrent connectivity.
• Employed Schaefer-400 cortical parcellation to define 400 functional nodes, preserving distributed cortical interactions.
• Mapped the derived task-state topology onto the SNN reservoir, using speech spike trains as input.

Main Results:

• Task-state fMRI revealed enhanced functional connectivity in auditory nodes compared to resting-state, indicating task-related activation.
• The fMRI-constrained SNN demonstrated improved speech recognition performance over baseline SNNs.
• The proposed topology enhanced the biological interpretability of the SNN model.

Conclusions:

• Task-state functional brain organization provides an effective topology prior for biologically inspired speech recognition models.
• Integrating neuroimaging data into SNN design offers a promising avenue for advancing artificial speech recognition capabilities.
• The study highlights the potential of using brain functional architecture to guide the development of more effective computational models.