Proprioception relies on a network of sensors within our muscles that send signals about limb position and movement to the brain. Yet, the method by which the brain integrates these signals has remained largely mysterious.

The EPFL team's research, detailed in the journal Cell, involved Alessandro Marin Vargas, Axel Bisi, and Alberto Chiappa as primary researchers, with Chris Versteeg and Lee Miller providing critical experimental data. The study utilized musculoskeletal simulators to understand the statistics of sensor distribution in muscles, simulating muscle spindle signals in the upper limb to compile a comprehensive movement dataset.

This dataset was then used to train numerous neural network models based on sixteen computational tasks, each representing a hypothesis about the proprioceptive system, including parts of the brainstem and somatosensory cortex. The analysis showed that models predicting limb position and velocity best replicated how the brain processes proprioceptive information, suggesting the brain prioritizes these inputs to comprehend body dynamics.

The findings underscore the utility of task-driven modeling in neuroscience, marking a departure from traditional methods focused on predicting neural activity. This approach could revolutionize our understanding of sensory processing and significantly advance neuroprosthetic technologies by providing more intuitive control over artificial limbs.

Research Report:Task-driven neural network models predict neural dynamics of proprioception.

Related Links
Swiss Federal Institute of Technology Lausanne
The Physics of Time and Space