Transcript: Decoding phantom limb movements from intraneural recordings

Alex: Welcome to another episode of ResearchPod. Today we're looking at a study that tackles a persistent problem in lower-limb prosthetics: how to give transfemoral amputees genuine, volitional control over their devices.

Sam: Most lower-limb prosthetics are passive — they don't respond to the user's intent. And while upper-limb research has made real progress on neural interfaces, lower-limb amputees have been largely left behind. The core problem is anatomical: the muscles you'd want to record from sit deep in the residual limb, well beyond what surface electrodes can reliably reach.

Alex: So the question this paper is asking is whether you can bypass surface sensors entirely and tap directly into the peripheral nerve to decode motor commands?

Sam: Exactly. The authors implanted transversal intrafascicular multichannel electrodes — TIMEs — directly into the sciatic nerve of two transfemoral amputees. The central claim is that the sciatic nerve retains rich, decodable motor intent for complex movements like knee flexion, ankle dorsiflexion, and toe extension — even years post-amputation. That's not a trivial result, because there's a real question about whether those efferent pathways remain organized and functional when they're no longer driving muscle.

Alex: And peripheral nerve recordings are notoriously noisy. How do they actually extract something usable from that signal?

Sam: That's the technical crux of the paper. They used a Spiking Neural Network — an SNN — as the decoder. The key distinction from a conventional neural network is that an SNN operates on discrete spike events rather than continuous voltage traces. The nerve's electrical output is already event-based, so the SNN is architecturally matched to the signal. Specifically, they used Leaky Integrate-and-Fire neurons, which accumulate incoming spikes over time and fire only when a threshold is crossed. That temporal integration is what lets the network extract timing information that a rate-coded decoder would discard.

Alex: So the SNN is treating spike timing as signal rather than noise — which is exactly what you'd want if the motor intent is encoded in the precise temporal structure of the discharge.

Sam: Right. And they didn't rely on nerve recordings alone. The second design decision is a hybrid signal integration: they concatenated the electroneurography recordings with intramuscular EMG from residual limb muscles. The hybrid approach meaningfully outperformed either modality in isolation, which tells you that the two sources are carrying complementary information rather than just redundant signal.

Alex: Did they see enough spatial selectivity in the nerve to distinguish between individual joints? That's where I'd expect the decoding to get difficult.

Decoding phantom limb movements from intraneural recordings

Cecilia Rossi, Marko Bumbasirevic, Paul Čvančara, Thomas Stieglitz, Stanisa Raspopovic, Elisa Donati, Giacomo Valle

Feb 8, 2026·5:13·doi·Original Paper

Intraneural recordingsPhantom limb movementsSpiking Neural Network (SNN)Transversal intrafascicular multichannel electrodes (TIME)Sensorimotor integrationIntraneural interfaces

Listen

5 min

Briefing

Research Question and Approach

Lower-limb amputees often face significant mobility challenges, and current prosthetic devices lack intuitive, volitional control and sensory feedback. This study investigates whether intraneural multichannel electrodes (TIMEs) implanted in the residual sciatic nerve can capture neural signals related to phantom limb movements. The researchers implanted four TIMEs in two transfemoral amputees and recorded neural activity while participants performed volitional phantom movements of the knee, ankle, and toes. They utilized a spiking neural network (SNN) to decode these movements, comparing its performance against conventional machine learning methods. Additionally, they explored the feasibility of using these same electrodes to provide sensory feedback and to record inter-muscular EMG signals.

Main Findings

The study successfully identified joint- and direction-specific neural modulation in the sciatic nerve, demonstrating that peripheral nerve interfaces can access motor information for multiple joints, including those lost after amputation. The SNN-based decoder significantly outperformed conventional classifiers (like SVM and MLP) in predicting attempted movements. The researchers found that the SNN's ability to leverage the temporal precision of spikes made it more robust and efficient for decoding neural intent. Furthermore, integrating inter-muscular EMG signals with the intraneural neural data further improved decoding accuracy. Finally, the researchers observed that motor and sensory maps showed minimal overlap, suggesting that afferent and efferent fibers are already segregated within the sciatic nerve at the thigh level, which facilitates bidirectional prosthetic control.

Why It Matters

This work provides a proof-of-concept for a single, implantable technology that can simultaneously record motor commands and deliver sensory feedback. By enabling direct neural control of prosthetics, this approach could significantly improve the functionality, embodiment, and user experience for individuals with limb loss. The use of neuromorphic hardware and SNNs suggests a path toward low-power, real-time, onboard prosthetic control, moving these systems closer to clinical viability.

Key Terms and Definitions

Intraneural Multichannel Electrode (TIME) — A penetrating electrode array implanted directly into the nerve fascicles to record or stimulate neural activity.
Spiking Neural Network (SNN) — A type of artificial neural network that processes information using discrete spikes, mimicking the event-driven nature of biological neural signals.
Phantom Limb Movement — The volitional attempt by an amputee to move the missing limb, which still generates neural motor commands in the residual nerves.
Electroneurographic (ENG) Signal — The electrical activity recorded from peripheral nerves, which contains information about motor commands and sensory feedback.
Inter-muscular EMG (imEMG) — Electrical activity recorded from muscles using electrodes placed within or near the nerve, providing information about muscle activation patterns.

Abstract

Limb loss causes severe sensorimotor deficits and often necessitates prosthetic devices, particularly in lower-limb amputees. Although direct neural recording from residual nerves offers a biomimetic route for prosthetic control, low signal amplitudes and challenges in nerve interfacing have limited adoption. Intraneural multichannel electrodes provide a potential solution by enabling access to motor signals from muscles lost after amputation. Here, we report intraneural recordings from two transfemoral amputees using transversal intrafascicular multichannel electrodes implanted in distal branches of the sciatic nerve. We identified multiunit activity associated with volitional phantom movements of the knee, ankle, and toes, exhibiting joint- and direction-specific modulation distributed across electrodes. A Spiking Neural Network-based decoder outperformed conventional methods in predicting attempted movements, with further gains achieved by integrating intraneural and intermuscular signals. Motor and sensory maps showed minimal overlap, indicating early segregation within the sciatic nerve. These findings pave the way for bidirectional, neurally-controlled prosthetic systems.

Decoding phantom limb movements from intraneural recordings | Cecilia Rossi et al. | ResearchPod

Decoding phantom limb movements from intraneural recordings: Research Paper Podcast