An AI Learned To Decode Phantom Limb Movements From Inside The Nerve: It Could Change Prosthetic Legs Forever

After amputation, the brain does not forget a missing limb. Motor pathways keep firing commands down nerves that no longer reach a leg, suggesting that movement-related neural pathways remain active even after limb loss. For the first time in human research, scientists have intercepted those signals and decoded intended movements across every joint of a missing leg, including the knee, ankle, and toes, and even movements tied to muscles that were surgically lost at the time of amputation.

A new study published in Nature Communications describes how researchers implanted tiny electrode arrays directly inside the sciatic nerve of two above-knee amputees and recorded nerve activity as each participant attempted phantom movements on command. A purpose-built artificial intelligence translated those signals into specific joint movements with greater accuracy than conventional decoding methods.

Lower-limb research in this area had lagged far behind work on the arms and hands. No prior study had decoded movement intentions from all three leg joint regions at once. Beyond that milestone, the work also points toward the possibility of a prosthetic leg that both receives neural commands and sends touch sensations back, through the same implanted device.

How Researchers Read Phantom Limb Signals From Inside the Nerve

Both participants had undergone above-knee amputation and used passive prosthetic devices. Surgeons implanted four multi-channel electrode arrays into the distal branch of the sciatic nerve in each participant’s residual thigh. Each array carried 14 active recording sites, giving the team 56 total data channels per person.

Participants were seated and asked to follow movement cues on a screen: flex the knee, extend it, flex the ankle, extend it, flex the toes, extend them. Each attempt lasted two seconds, followed by a two-second rest. No movement could actually occur, but the nerves kept firing.

Across both participants, the electrodes picked up clear activity tied to each intended movement. In the first participant, 91% of recording sites responded to at least one phantom movement. Many sites fired selectively for specific joints or motion directions. Toe-related signals were detected even though the muscles controlling toe movement had been lost at amputation.

That last point matters. Conventional surface sensors on the skin cannot reach signals from muscles that no longer exist. Reading activity from inside the nerve, upstream of where those signals would have traveled, gave researchers access to motor intent that external sensors struggle to capture.

Decoding Phantom Limb Movements With a Brain-Inspired AI

To make sense of the neural data, the team built a spiking neural network, an AI that processes information in discrete pulses the way real neurons do. Standard machine learning models work with averaged or continuous signals; this system was designed to exploit the precise timing of individual nerve spikes instead.

Tested against two conventional approaches, the spiking network outperformed both. For the first participant, working with six movement classes covering flexion and extension of the knee, ankle, and toes, the best configuration reached roughly 64% accuracy when nerve and residual muscle signals were combined. For the second participant, working with four movement classes, accuracy exceeded 72%.

An unexpected bonus came from the electrode placement itself. Sitting inside the thigh between surrounding muscles, the arrays also picked up low-frequency electrical signals from nearby residual muscle tissue. Adding that data to the decoder improved accuracy further, with no additional surgery required.

One Phantom Limb Implant That Can Both Read Movement and Restore Touch

A separate finding changed how the team understood the nerve’s internal structure. Most electrode sites that detected motor signals were physically separated from the sites that, when stimulated, produced touch sensations in the phantom limb. Motor and sensory fibers appear to occupy largely distinct regions of the sciatic nerve at the thigh level, before the nerve splits into its major branches.

Prior research by several of the same authors had already shown that stimulating these electrodes could restore sensations of touch and pressure in leg amputees. A single implant that reads intended movements through one group of sites and delivers sensory feedback through another could enable a two-way neural connection not available in current commercial prosthetics.

Lower-limb amputation accounts for roughly 69% of all limb loss. People who lose a leg above the knee typically face reduced walking speed, higher energy expenditure, and significant loss of independence. Most commercial prosthetic legs rely on passive mechanisms with no direct link to the user’s nervous system.

Researchers have now shown that the signals the brain sends to a missing leg carry enough information to distinguish between multiple intended joint movements, and that an AI designed to process data the way neurons do is better equipped to read them than conventional tools.

---

Disclaimer: This article is based on a published peer-reviewed study. The research involved only two participants, and all movement decoding was performed offline under controlled conditions. Findings should not be interpreted as a clinically available treatment or device. Further research with larger samples and real-time testing is needed before these methods could be applied in a clinical or commercial setting.

---