A redundant encoding algorithm for artificial sensory information speeds learning and improves multisensory-guided navigation

Intracortical microstimulation (ICMS) directly influences cortical activity, serving as an artificial sensory stimulus to guide precise control of prosthetic limbs. However, only a small portion of sensations triggered by ICMS are proprioceptive, meaning they describe the position and movement of the body. A learning-based approach to encoding an artificial proprioceptive signal addresses this issue.

For instance, animals trained with multi-channel ICMS alongside natural vision learn to both decode the ICMS signal to locate an invisible target and integrate artificial sensation with natural sensation. Yet, the drawback of this learning-based method is the need for training. Researchers hypothesized that modifying the algorithm used to encode artificial sensory information could both shorten learning time and enhance plateau performance on ICMS-guided navigation.

To verify this hypothesis, eight mice underwent training on a sensory-guided navigation task: locating a target in a training cage, with the target's position encoded through either a red circle (vision), multi-channel ICMS, or a combination of both vision and ICMS. ICMS was encoded using two distinct algorithms: sparse (employing fewer simultaneously stimulating electrodes) or redundant (utilizing more simultaneously stimulating electrodes).

Researchers discovered that encoding ICMS with a redundant algorithm accelerated the learning process of the ICMS signal compared to sparse encoding. Moreover, mice guided by redundant ICMS ran faster and completed trials more swiftly than those guided by sparse ICMS. Additionally, redundant encoding facilitated multisensory integration of ICMS with natural vision, resulting in improved success rates, path efficiency, movement speed, and movement time.

The researchers concluded that optimized ICMS encoding algorithms could overcome the existing limitations of learning-based sensory encoding, enabling learning and integration of artificial sensory information into existing sensorimotor neural circuits. This advancement paves the way for restoring both sensory and motor streams of information flow in damaged sensorimotor systems.