Temporal interference stimulation of peripheral nerves induces functionally diverse limb movements revealed by automated pose estimation and unsupervised behavioral analysis.

Peripheral nerve stimulation can help restore limb movement after paralysis and enable advanced rehabilitation technologies; however, current extraneural interfaces are typically limited by low fascicle selectivity and laborious functional evaluation. This study has developed an extraneural peripheral nerve interface with high fascicle selectivity, and an AI-facilitated video analysis pipeline for assessing limb movement during neuromodulation. This was achieved by deploying temporal interference stimulation (TIS) in a high-density nerve cuff electrode and by using machine learning algorithms for automated pose estimation and unsupervised behavioral analysis to evaluate movement selectivity and diversity. Using this unbiased semi-automated analysis revealed that TIS elicited more selective motor responses than standard biphasic stimulation, as evidenced by the formation of 1.75 times more distinct movement clusters and behavioral syllables. Furthermore, logit link beta regression modeling showed that TIS had a significantly higher positive effect on movement selectivity (β = 2.75, p < 0.005) compared to biphasic stimulation. Our statistical and machine learning-based analysis provides a computational and objective pipeline for quantifying complex motor outcomes in neuromodulation research. The results suggest that extraneural TIS can be used to generate individually targeted and functionally diverse limb movement patterns and offers a promising approach for neurorehabilitation applications, including restoring movement to individuals living with spinal cord injury.