Neural NoiseBCI Annual Review — 2006
1 January–31 December 2006
Introduction
2006 was the year BrainGate entered the scientific literature in authoritative form. On July 13, 2006, the Nature paper “Neuronal ensemble control of prosthetic devices by a human with tetraplegia” appeared, authored by Hochberg, Serruya, Friehs, Mukand, Saleh, Caplan, Branner, Chen, Penn, and Donoghue, presenting the first peer-reviewed results from the Cyberkinetics BrainGate clinical trial. The paper reported on two participants: Matthew Nagle (the first implant, June 2004, 57 sessions) and a second participant implanted at the University of Chicago (April 2005). Nagle’s data demonstrated that a human with cervical spinal cord injury could control a 2D computer cursor with sufficient precision to open simulated email, operate a television, draw circular shapes, play “neural Pong,” and command a prosthetic hand — all using motor cortex signals recorded through 96 implanted microelectrodes years after his spinal cord injury. The publication in Nature, with its associated media coverage and editorials, marked the BCI field’s definitive arrival in mainstream biomedical consciousness. The same July 13 issue of Nature also contained the Santhanam, Ryu, Yu, Afshar, and Shenoy paper “A high-performance brain–computer interface” from Stanford, demonstrating that premotor cortex plan-activity decoding in monkeys could achieve 6.5 bits per second — a fourfold improvement over prior systems — by focusing on the delay-period preparatory activity rather than movement-epoch firing rates.
The dual appearance of these two papers in the same Nature issue was not coincidental — it represented the field’s twin frontiers: the clinical demonstration of feasibility in a human patient, and the pre-clinical push toward information rates practical enough for real-world use. Together they defined 2006 as the year intracortical BCI transcended proof-of-concept and began confronting the question of performance. The Hochberg et al. paper was candid about limitations: Nagle’s control was “considerably less than that of an able-bodied person using a manually controlled computer cursor,” signal quality had declined after 6.5 months, and the system remained bulky and dependent on daily recalibration. The Shenoy paper showed what would be needed to make such systems truly practical. The conversation the field would have for the next decade — between those focused on safety/feasibility and those focused on performance/information rate — was defined in these two July 2006 publications.
Beyond the dual Nature papers, 2006 saw the BCI Competition III results formally reviewed and published in IEEE Transactions on Neural Systems and Rehabilitation Engineering (vol. 14, no. 2), providing the most comprehensive benchmarking of decoding algorithms to date, with papers from each winning team. The Seattle ECoG-BCI group (Rao, Ojemann, and colleagues at the University of Washington) published their extended ten-patient series in the same journal issue, confirming that ECoG-based one-dimensional cursor control was rapid and reliable, and demonstrating for the first time that epidural (outside the dura) ECoG recordings could also support BCI control — a potentially less invasive alternative to subdural placement. The Neurochip BCI from Fetz’s group at the University of Washington — an autonomously operating interface that could record from single motor cortex neurons and stimulate another site simultaneously, forming a closed-loop artificial synapse — was described in a 2006 paper, opening the path toward bidirectional BCIs that both read from and write to the nervous system.
The neuromodulation field continued to expand. DBS was being used at an increasing number of centers for dystonia (under the 2003 HDE), and the literature on optimal programming parameters for PD subthalamic DBS was maturing. Closed-loop DBS concepts — in which stimulation parameters would be automatically adjusted based on recorded local field potentials from the DBS electrode itself — were being conceptually advanced, with early papers from Bergman (primates) and Brown (humans) describing the detection of pathological beta oscillations in the STN as a potential biomarker for adaptive stimulation. This work presaged the “adaptive DBS” systems that would eventually reach clinical trials a decade later.
Timelines
January–March. Hochberg, Donoghue, and co-authors were completing revisions of the BrainGate Nature manuscript with the editorial team. Stanford’s Santhanam et al. manuscript was simultaneously in revision at Nature. At Wadsworth, the P300 speller in BCI2000 continued to be refined and deployed in ALS patient evaluations at multiple sites. The BCI Competition III review paper by Blankertz et al. was in preparation and the individual winning-team papers were being finalized for the June 2006 issue of IEEE Trans. Neural Systems and Rehab. Eng. Cyberkinetics was in discussions about additional BrainGate trial sites, with Leigh Hochberg at MGH/Harvard actively screening new participants for the ongoing pilot trial.
April–June. The BCI Competition III review and winning-team papers appeared in IEEE Trans. Neural Systems and Rehab. Eng. 14(2), June 2006. Leuthardt and Ojemann’s second ECoG series (ten patients, University of Washington) appeared in the same issue. The Neurochip BCI paper by Jackson, Mavoori, and Fetz appeared in IEEE Trans. Neural Systems and Rehab. Eng. in June 2006, describing a wearable, autonomous BCI chip that recorded single-unit motor cortex activity and delivered stimulation to another neural site without external computer involvement, sustained over weeks in unrestrained monkeys. This was the first demonstration of a truly self-contained, implantable BCI without a tethered external computer. Krishna Shenoy was a speaker at the June 2006 Summer Institute for BCI Research, presenting the forthcoming 6.5 bps results to the community.
July–September. The dual Nature papers appeared on July 13, 2006, generating significant press coverage. The BBC headline “Brain chip reads man’s thoughts” was characteristic of the media response to Hochberg et al., though scientists took care to clarify that the system decoded motor intent rather than reading arbitrary thoughts. Donoghue and Hochberg gave press briefings emphasizing the clinical importance of demonstrating that long-standing paralysis did not eliminate usable motor cortex signals. The Shenoy paper was simultaneously published; its emphasis on information rate and the practical target of 15 words per minute as a threshold for clinical utility attracted attention from assistive technology researchers and clinicians who immediately recognized that comparison point. Andrew Schwartz spoke at the Society for Neuroscience summer satellite meeting on his ongoing primate work; the first public discussions of the monkey “self-feeding” experiments were being heard in the research community.
October–December. The Society for Neuroscience annual meeting in Atlanta (October 2006) featured what many described as a BCI “coming out party” — multiple well-attended symposia, hundreds of related posters, and exhibits from Cyberkinetics demonstrating BrainGate to a broad neuroscience audience. Closed-loop DBS papers from Brown’s group (Peter Brown, University of Oxford) describing beta-band LFP oscillations in the STN as markers of motor dysfunction in PD patients attracted attention as potential biomarker targets for adaptive DBS. Cyberkinetics was exploring funding options and beginning to show signs of the financial pressure that would eventually lead to its restructuring. A third BrainGate participant — a woman with brainstem stroke — received an implant in late 2005 or early 2006 (participant S3), and her data were being accumulated during this period; she would later be reported to have useful neural signals more than 1,000 days post-implant.
Trends
BrainGate in Nature: Clinical Feasibility Enters the Mainstream Record
The Hochberg et al. 2006 Nature paper settled definitively what had previously been known only to those following conference presentations and press accounts: a human being with chronic tetraplegia could use a 96-electrode intracortical BCI to control a computer cursor in real time. The paper’s primary contribution was not simply the demonstration — which Nagle’s trial had provided informally since 2004 — but the rigorous documentation of neural signal properties, decoder performance metrics, and the time course of signal evolution over the 57-session trial. It confirmed that motor cortex neurons in a person with years of cervical SCI retained directional tuning and temporal modulation suitable for BCI decoding, and quantified the information content of the BrainGate signal. The companion editorial in Nature described the result as opening “a new era in our understanding of the human brain,” and it catalyzed funding agency interest in the field.
Stanford’s Information Rate Breakthrough: Plan Activity and Clinical Targets
The Santhanam et al. paper in the same Nature issue as Hochberg et al. advanced a different but complementary argument: raw cursor control was not sufficient; what mattered was how fast and accurately a BCI could enable a user to select targets. By leveraging the brief planning-period neural activity in premotor cortex before a reach, rather than the movement-phase activity itself, the Stanford group extracted information at 6.5 bits per second — equivalent to approximately 15 words per minute in a spelling application. This was the first time a BCI system had demonstrated information rates competitive with existing assistive communication alternatives such as eye-gaze systems. The paper also introduced the “neural population” framework for high-dimensional decoding in a clinically oriented context, influencing the direction of multiple groups.
ECoG Confirmed as a Reproducible BCI Modality
The Seattle group’s ten-patient ECoG series, published in June 2006, was important for establishing reproducibility. Where the 2004 Leuthardt paper had four patients, the expanded series showed that all subjects who attempted closed-loop BCI achieved control, with accuracies between 73 and 100%. The finding that control was achievable using motor imagery, sensory imagery, speech imagery, and even auditory cortex activation gave ECoG an unusual versatility — patients who could not perform motor imagery (e.g., those with extensive motor cortex injury) might still find a usable control signal. The additional finding that epidural recordings (from outside the dura) could support BCI control further reduced the surgical risk profile, suggesting that ECoG-based BCIs might be implanted through simpler procedures than subdural placement.
The Autonomous Implantable BCI: Neurochip
Fetz’s group at the University of Washington described the Neurochip in 2006: a 2.5 g, watch-case-sized device that could be mounted on a monkey’s head and autonomously record single-unit motor cortex activity while delivering programmable stimulation to another site — all without external cable connection. Sustained operation over weeks in unrestrained monkeys demonstrated both biocompatibility and reliable long-term recording. The Neurochip represented the logical endpoint of a trend toward miniaturization and implantability that all BCI groups recognized would be necessary for clinical translation: a device that the patient could wear or have fully implanted, without daily calibration by a technician and without a hospital-room-sized equipment cart. While the Neurochip itself was not a complete BCI system, its architecture anticipated the fully implantable wireless BCI designs that would emerge over the following decade.
Adaptive DBS: Closing the Loop on Neuromodulation
Peter Brown’s group at the University of Oxford, working with Bergman’s group in primates, was demonstrating that the pathological beta-band (13–30 Hz) local field potential oscillations recordable through standard DBS electrodes in the subthalamic nucleus of Parkinson’s patients were highly correlated with motor symptoms, and that these oscillations were suppressed by therapeutic stimulation. This work suggested that DBS could, in principle, be made “adaptive” — automatically adjusting stimulation in response to ongoing LFP biomarkers rather than delivering constant stimulation. The implications for BCI were significant: DBS electrodes were already in tens of thousands of patients, and if their LFP recordings could be used as control signals, a ready-made population of implanted patients could be recruited for BCI studies. The idea of using DBS LFPs for BCI communication would be actively explored in subsequent years.
Suggested Titles
- The July 13 Manifesto: Nature, BrainGate, and Stanford Define the Field’s Dual Agenda
- From Lab Notebooks to Peer Review: The Year BCI Proved Itself in Print
- Six-Point-Five Bits and Four Participants: BCI Performance and Clinical Reality in 2006
- ECoG Confirmed, Neurochip Launched, Adaptive DBS Conceived: The Architecture of Tomorrow
- Neural Pong and 6.5 bps: Measuring a Revolution in Its First Clinical Year