From Skin to Cortex: End-to-End Spiking Neural Network Simulation of Tactile Information Flow

Abstract

Autonomous systems and neuroprosthetic devices demand real-time tactile processing under strict energy and latency constraints. Designing these systems using neuromorphic principles, where communication is event-based and node activity is sparse, could improve their speed and energy efficiency. Here we present an end-to-end spiking neural network model of the ascending tactile pathway, from mechanoreceptors in the skin in humans (or whisker follicles in rodents) to cortical neurons, that operates in an event-driven neuromorphic fashion. The model comprises distinct anatomical stages, (1) three types of mechanoreceptor afferents, (2) trigeminal ganglion, (3) brainstem, (4) thalamic nucleus, and (5) three cortical layers, connected in a feed-forward hierarchy. We demonstrate the models responses to both rodent whisker deflection and human fingertip skin displacement, using information-theoretic analysis, pairwise correlation, and stimulus decoding at each layer. Our results show that tactile information is efficiently encoded and transformed at each stage: stimulus features are represented with high fidelity and reduced redundancy as signals ascend. Notably, simple linear or Bayesian decoders can reliably classify stimulus features from single-neuron activity in the thalamus and cortex for low-noise inputs, highlighting the emergence of robust neural representations. This open-source model is the first to include the mechanosensory periphery in a full tactile pathway simulation, enabling researchers to study how perturbations at any stage affect tactile encoding. Moreover, the network is well-suited for deployment on low-power, real-time neuromorphic hardware, facilitating the development of multi-layer signal processing and tactile navigation algorithms.