Keith Thomas can feel his dog: Feinstein's three-year double neural bypass lands in Nature Medicine

By Inside BCI · July 18, 2026

Keith Thomas of Massapequa, New York, dove into a swimming pool on 18 July 2020 and broke his neck. The paper his implant produced went up on the cover of Nature Medicine on 16 July 2026, two days shy of the six-year anniversary of the accident that left him a complete C4-sensory, C5-motor tetraplegic. In the intervening years, a Northwell Health surgical team led by Ashesh Mehta and Netanel Ben-Shalom opened his skull for fifteen hours in March 2023 and slid five microelectrode arrays into his brain. Two of the arrays went into the hand region of his primary motor cortex, three into the hand region of his primary somatosensory cortex, wired to a hybrid stimulation system that Chad Bouton, who spent eighteen years at Battelle Memorial Institute before joining Feinstein around 2018, had been building for over a decade. Thomas was forty-two at enrolment. He is forty-eight now. He can feed himself, drink from a cup, scratch his own face, wipe his own eyes, feel the touch of his sister Michelle’s hand, and pet the family dog, a Malshipoo named Bow who arrived just as the arrays did.

“Being able to feel my sister’s hand, to pet my dog and feel her fur,” Thomas said in the Feinstein Institutes’ 16 July release, “these experiences that the injury took away have been restored. But beyond the study sessions, I can now scratch my face, wipe my eyes independently. The technology has given me back both connection and sense of self.”

The paper is the primary three-year follow-up on the Feinstein “double neural bypass” trial (ClinicalTrials.gov NCT03680872, FDA IDE G170200, Northwell Health IRB 17-0840), an FDA-investigational device trial that has been running since 2018 and enrolled Thomas as its first-in-human participant in 2021. Santosh Chandrasekaran and Sarah K. Wandelt are the co-first authors; Chad Bouton is the corresponding senior author. The author list runs to twenty-six across a multi-institution consortium anchored at Feinstein Institutes and Northwell Health, with contributors from the Zucker School of Medicine at Hofstra/Northwell, Johns Hopkins, Johns Hopkins Applied Physics Laboratory, Baylor College of Medicine, Washington University in St. Louis, and additional partner institutions. The paper occupies pages 2591 through 2601 of Nature Medicine Volume 32 and appears on the July 2026 cover of the journal. The DOI is 10.1038/s41591-026-04498-0.

What the paper actually reports

The “double” in “double neural bypass” refers to two closed loops running simultaneously through Keith Thomas’s damaged spinal cord. The motor loop reads intent to move from an implanted intracortical array in his primary motor cortex, feeds the raw neural signals to a long short-term memory recurrent neural network trained on 100-millisecond bins of broadband cortical activity, translates the decoded intent into stimulation commands, and delivers those commands to two output channels running in parallel: a set of neuromuscular electrical stimulation patches on the surface of his forearm that drive his own muscles, and a custom transcutaneous spinal cord stimulation patch worn over the skin at his cervical spine that targets the C5 to C8 spinal roots with a 10-kilohertz biphasic waveform. His own hand contracts. The sensory loop reads mechanical pressure from force sensors embedded in a three-dimensionally printed active orthosis worn on his right hand, converts that pressure signal into a pattern of intracortical microstimulation delivered through up to sixteen electrodes in the somatosensory cortex arrays, and pairs the cortical stimulation with vibrotactile stimulation on the skin and with the same transcutaneous spinal cord stimulation carrying the motor signal. Thomas perceives touch on his own hand.

The specific measured outcomes are worth listing precisely. Over a 35-week structured intervention block, Thomas’s right-arm elbow flexion strength rose by 86 per cent and his left arm rose by 62 per cent, with the right-arm improvement significant at P equals 0.0003 and the left arm at P equals 3.8 times ten to the minus fifth. On a task requiring him to grasp hollow eggshells without breaking them, the reinforcement-learning force-control agent hit an 87 per cent success rate. Motor decoding accuracy, from the locked decoder without retraining, held at 84.6 per cent sustained across five months. Tactile sensation returned to his right radial wrist after approximately 25 weeks of combined stimulation, with Semmes-Weinstein monofilament testing showing detection of forces as low as 10 grams. And here is the finding that most distinguishes the paper from what has come before it in the hybrid brain-computer interface literature. When the stimulation was switched off, Thomas kept the sensation. Bouton’s on-record quote in the Feinstein release: “Remarkably, in a recent follow-up, it was found that these gains were still present after more than two years. This is incredibly encouraging.”

The paper attributes the durable persistence of function after stimulation ends to network-level neuroplastic changes in Thomas’s own somatosensory cortex, with a linear mixed-effects model showing a significant increase in modulation depth on the stimulated electrodes (beta equals 0.20, P equals 2.8 times ten to the minus eighth). This is a materially different sensory-restoration mechanism from what has historically come out of the Pitt intracortical microstimulation programme that Inside BCI covered in the 15 July 2026 Science Translational Medicine paper on Nathan Copeland. Copeland’s percepts are cortically evoked by microstimulation delivered during the stimulation session. Thomas’s percepts are, after twenty-five weeks of paired stimulation, biologically re-processed touch signals arriving through his own peripheral nervous system, entering a somatosensory cortex that has been retrained by the cortical-mirroring protocol into re-engaging with them. The distinction matters for the commercial and clinical implication space that the paper opens up, because a device that produces durable, take-home biological recovery is a fundamentally different product from a device that must run continuously to preserve function.

What is not new, and what genuinely is

Chad Bouton has spent his career on this specific architecture. Before he joined Feinstein Institutes around 2018 and eventually took over as Center Head of the Center for Bioelectronic Medicine and Vice President of Advanced Engineering at Northwell Health, he spent eighteen years at Battelle Memorial Institute in Columbus, Ohio. His first-in-human trial there was Ian Burkhart, a twenty-four-year-old man who broke his own neck in a diving accident and received a Utah microelectrode array in April 2014 in a surgery performed by Ali Rezai at Ohio State’s Wexner Medical Center. Bouton was first author on the resulting 2016 Nature paper. Burkhart, with Bouton’s Battelle NeuroLife system, opened and closed his own paralysed hand, poured water from a bottle, swiped a credit card, and played Guitar Hero. Later Battelle work on Burkhart, published in the peer-reviewed literature and in Battelle’s own press materials, added a haptic vibration armband on his upper arm that decoded subperceptual residual touch signals from his motor cortex and translated them into skin buzzes so that he could sense what his hand was doing.

The Battelle NeuroLife system was, in structural terms, a motor bypass with sensory substitution. The Feinstein double neural bypass is a motor bypass with a genuine sensory bypass and, on top of it, a transcutaneous spinal cord stimulation layer that Battelle did not have. What is genuinely new in Thomas’s system, then, is not the idea of pairing an intracortical motor decoder with peripheral muscle stimulation and calling it a bypass. That was Bouton’s own 2014 idea, and it worked on Burkhart. What is new is closing the sensory loop through the somatosensory cortex rather than through a vibrating armband on the upper arm, adding transcutaneous stimulation at the level of the spinal cord to drive the plasticity, and documenting that the composite intervention produces persistent post-stimulation biological recovery in a person with a complete tetraplegic injury.

Inside BCI is not using unqualified “world’s first” language on this paper. Bouton is quoted in the Feinstein release saying his team “accomplished something that’s never been done before,” and taken narrowly, that is defensible. The narrow claim that survives adversarial fact-check is that Keith Thomas is the first person with an ASIA Impairment Scale A complete cervical tetraplegic injury to have both movement of his own hand and cortically routed touch sensation from that hand restored by a single integrated implanted system in a peer-reviewed publication, with the recovery persisting more than two years after stimulation ended. Every load-bearing word in that sentence excludes a specific piece of prior art. “His own hand” excludes the robotic-arm demonstrations from BrainGate participants Cathy Hutchinson, Nathan Copeland, and Jan Scheuermann. “Cortically routed touch” excludes the Battelle NeuroLife haptic-armband sensory substitution. “Single integrated implanted system” excludes the strategy of stacking independent devices. “Complete tetraplegic injury” excludes cohorts with residual function. “Peer-reviewed publication” excludes prior press releases and demonstrations. The paper’s contribution is precise, and describing it precisely is more useful to the reader than repeating Feinstein’s marketing.

The other person in the story

Kevin J. Tracey, Feinstein’s president and chief executive, is the third named institutional voice on the release. Tracey has been shaping Feinstein Institutes for a quarter century and is himself a foundational figure in the bioelectronic medicine field that Bouton’s Center now leads. His on-record framing at the launch is worth quoting directly: “Dr. Bouton and his team have significantly advanced the fields of bioelectronic medicine and brain-computer interfaces. By combining brain-computer interface technology with high-precision neuromodulation they have opened new therapeutic possibilities for restoring function after injury to the brain and nervous system.”

Thomas is not the only person the trial has enrolled. The Feinstein release describes a subsequent protocol in which Thomas’s implant was used to help a second participant with paralysis move a hand, while Thomas himself felt sensations from the second person’s touches on his own fingertips. The interhuman-bypass sub-study is a separate protocol from the primary Nature Medicine paper and Inside BCI is not reporting on it here beyond noting its existence. It is a signal that Feinstein is treating Thomas’s specific brain, three years and four months after implantation, as continuing research infrastructure.

What it looks like now, and what it does not yet do

The system is not portable. Thomas comes to the Feinstein lab. The AI decoder runs on a gaming computer connected by an HDMI cable to a skull-mounted amplifier. The forearm patches and the transcutaneous spinal patch and the active orthosis all have to be applied and tuned each session. The 2024 press cycle around this same trial, when Chad Bouton spoke to TIME, framed the ambition as “eventually to condense this to portable form.” That has not happened yet, and the paper does not claim it has. The non-invasive sibling of this system, the transcutaneous-only stroke-recovery product line that Bouton’s team started developing at Battelle, has spun out to a company called Activate Neuro, which Inside BCI covered on 25 June 2026. The invasive cortical version that Thomas received has no equivalent commercial path yet. Neither the Feinstein release nor the paper discloses adverse events verbatim; Inside BCI is not asserting either their presence or their absence, and safety data will be in the ClinicalTrials.gov listing for the NCT03680872 trial.

Where this sits alongside the Copeland paper

The 15 July 2026 Science Translational Medicine paper from Robert Gaunt and Charles Greenspon at Pitt and the University of Chicago, and the 16 July 2026 Nature Medicine paper from Chad Bouton at Feinstein, are cousin stories that arrived one day apart and share almost no methodology. Pitt is purely cortical, purely stimulation-side, purely somatosensory, five participants, ten combined years of implant time, one hundred sixty-eight million pulses, no peripheral or spinal stimulation layer, no motor decoding integrated for own-limb control. Feinstein is hybrid cortical plus transcutaneous spinal plus peripheral muscle stimulation, five microelectrode arrays across motor and sensory cortex in one person, a single participant followed for more than three years, closed motor and sensory loops. The two papers together mark 16 July 2026 as the day the sensory-restoration limb of the brain-computer interface field materially expanded, in two different ways, from two independent programmes. Neither invalidates the other. The commercial roadmaps they imply are very different.

What to watch

Watch whether Feinstein enrols a second and a third participant on the primary NCT03680872 protocol within the next twelve months. The 2023 press cycle around the first surgery framed the trial as designed for up to three participants. Whether Thomas remains the only person publicly documented on the primary protocol, or whether a second surgery follows on the strength of the Nature Medicine cover feature, is the specific clinical-scaling signal to track. Second-participant data would answer the durability-versus-idiosyncrasy question the paper implicitly raises.

Watch whether Chad Bouton files a device sponsor’s Investigational Device Exemption amendment to move the system out of the lab and into home use. The 2024 TIME framing of “eventually portable” is now, with the Nature Medicine cover in hand, a specific regulatory milestone Feinstein can pursue. A move toward home-use IDE would be the operational proof point that the persistent-recovery finding survives outside the controlled laboratory environment.

Watch how the two same-week papers land with the US commercial invasive brain-computer interface cohort. Neuralink’s N1 array has more channels and a thinner form factor than the Utah-style arrays Thomas is implanted with, but Neuralink has not yet demonstrated somatosensory feedback in a human. Paradromics’ Connexus is motor-decode only in the University of Michigan first-in-human. Synchron’s Stentrode is endovascular and does not do intracortical stimulation. Precision Neuroscience’s Layer 7 is a surface electrode array. Whether any of the four commercial programmes files a sensory-feedback protocol amendment in response to the Feinstein durability data will indicate whether the “persistent post-stimulation biological recovery” story has traction beyond the academic literature.

Watch Keith Thomas. He is forty-eight years old, back home in Massapequa, and continuing to feed himself and drink from cups and pet his dog. The Feinstein trial is registered as active on ClinicalTrials.gov. The five arrays in his brain, the transcutaneous stimulation patches, and Bow the Malshipoo are all still going.

Sources

news Feinstein-Institutes Northwell-Health Chad-Bouton Keith-Thomas Nature-Medicine double-neural-bypass hybrid-neuroprosthesis tetraplegia cervical-spinal-cord-injury transcutaneous-spinal-cord-stimulation cortical-mirroring somatosensory-cortex neuroplasticity Battelle-NeuroLife Ian-Burkhart