Bionic prostheses and neuroprosthetics

Mechanics stopped being the bottleneck in prosthetics some time ago: actuators, batteries and multi-articulating composite hands all exist. The bottleneck is control. The device has to learn what the person intends to do, and intent lives in the nervous system, which is hard to reach.

The signal and its ceiling

The standard source is surface electromyography: electrodes on the residual limb pick up the summed potential of muscle fibres beneath. Skin and subcutaneous fat act as a volume conductor and a low-pass filter, so one muscle’s signal smears over several centimetres, neighbouring muscles contribute crosstalk, and the electrode records a mixture. Hence the ceiling. Two or three antagonistic patterns can be separated reliably; distinguishing the intent to flex the index finger from the intent to flex the ring finger, from skin over the forearm, generally cannot be done. Machine-learning classifiers widen the set of recognised postures but do not create information the signal lacks, and they degrade with electrode shift, sweat and changes in arm posture — not a software defect but a change in the physical problem itself.

Reinnervation as an amplifier

Targeted muscle reinnervation attacks the problem differently. The severed nerves of the residual limb — median, ulnar, radial — are surgically rerouted into denervated regions of a remaining muscle, typically segments of pectoralis major. The muscle becomes a biological amplifier: the motor command to close the hand, still travelling down the median nerve, now contracts its own patch of chest muscle, which is easy to read from the skin. The number of independent signals becomes the number of nerve branches that could be given separate muscle territory. A clinically valuable side effect is targeted sensory reinnervation: afferents grow into the overlying skin, creating a patch where touch is felt as touch on the missing hand.

Attachment

A socket transfers load through soft tissue that was never built for it, which produces abrasion, pistoning and lost positional accuracy. Osseointegration — a percutaneous implant anchored in bone — removes that and adds osseoperception, since vibration and force reach the skeleton directly. It buys two new constraints. One is mechanical: the exit site is a stress concentrator, and bone remodels around the altered load path. The other is infectious. A permanent skin opening, where epithelium cannot close around the implant, is an open route for colonisation, and stoma infection remains the leading cause of revision.

What is missing

The deficit is afferent. An intact hand reports position, force and slip continuously; a prosthesis reports nothing, so the user drives it visually, slowly and with high cognitive load, and drops the object the moment they look away. Attempts to close the loop — intrafascicular stimulation of the residual nerve, electrotactile and vibrotactile substitution — produce sensation but not the resolution or the natural somatotopic placement of the real thing. The absence of feedback also explains prosthesis abandonment rates: a device that demands constant attention gets left in a drawer.