Neural NoiseBCI Annual Review — 2007
1 January–31 December 2007
Introduction
2007 was marked by institutional transition, clinical consolidation, and deepening pre-clinical work in neural prosthetics. The year opened with Cyberkinetics Neurotechnology Systems, the commercial sponsor of the BrainGate pilot trial, facing significant financial pressure. Despite the landmark 2006 Nature paper and the scientific success of the trial to that point, the capital markets for medical devices in the mid-2000s were challenging, and Cyberkinetics — which had raised approximately $15 million and spent roughly $10 million by 2005 — struggled to attract the additional $40–50 million Donoghue and CEO Timothy Surgenor had estimated would be needed to commercialize and expand the BrainGate system. By 2007, in response to business pressures, Cyberkinetics shifted its strategic focus toward other medical device products, effectively winding down its direct investment in the BrainGate BCI trial. Clinical research continued through NIH grants and subcontracts, but the commercial pathway that had driven the 2004 IDE and the ambitious five-patient trial structure was narrowing. This financial difficulty foreshadowed the eventual withdrawal of Cyberkinetics from BCI research entirely by 2008–2009 and the transition to an entirely academically run trial under the BrainGate2 framework.
Despite these commercial difficulties, the scientific productivity of the field accelerated. In a move that demonstrated the BrainGate system’s potential for long-term utility, BrainGate participant S3 — a woman with a brainstem stroke who had been implanted in 2005 or early 2006 — was continuing to provide useful neural recordings well past the one-year mark, accumulating data that would eventually demonstrate reliable cursor control more than 1,000 days post-implant. This was a significant counter-narrative to the signal degradation story that had dominated discussion since the Nagle trial: some participants showed far more durable recordings than others, and understanding the determinants of recording longevity became a high priority. Matthew Nagle, the first BrainGate participant, passed away on July 24, 2007, from septicemia — a complication unrelated to the BrainGate implant — ending his extraordinary contribution to the field at age 27.
On the pre-clinical front, Andrew Schwartz’s group at the University of Pittsburgh published in Nature (May 2008) the results of experiments in which monkeys controlled a multi-joint robotic arm to self-feed — reaching, grasping, and delivering marshmallows and fruit to their mouths — but the underlying experiments were conducted in 2006–2007, with the paper submitted and in review through much of 2007. This work, demonstrating seven-degrees-of-freedom arm control from motor cortex recordings, would define the state of the art for BMI-controlled robotic limbs at the close of the decade. Meanwhile, Nicolelis’s group at Duke was pursuing experiments in which monkeys controlled a virtual avatar on a screen and even, in one remarkable experiment transmitted to Japan via internet, controlled a bipedal robot’s locomotion in real time using motor cortex oscillations — an early demonstration of the “Walk Again Project” conceptual framework Nicolelis would articulate publicly in subsequent years.
Non-invasive BCI research was maturing rapidly. The ASHA Leader published a detailed clinical review of P300 BCI technology in September 2007, specifically targeted at speech-language pathologists and clinical practitioners, emphasizing that BCI2000-based P300 spellers were being used in more than 80 research centers and could serve as a viable communication alternative for patients with ALS and other progressive neurological conditions. EEG-BCI research was increasingly attending to the needs of non-expert users and patients who could not train extensively — leading to interest in approaches requiring minimal calibration, and in the use of co-adaptive algorithms that improved in parallel with user skill. The SSVEP (steady-state visual evoked potential) paradigm, which required no training and could achieve high selection rates with modest accuracy, was gaining attention as a complement to the P300 and motor imagery paradigms for rapid character selection.
Timelines
January–March. Cyberkinetics’ financial restructuring was underway, with the company exploring strategic alternatives while continuing to support active trial participants through subcontracts. Hochberg and Donoghue maintained the BrainGate clinical research program through NIH funding, with BrainGate participant S3 (brainstem stroke) continuing to provide high-quality neural recordings. The Walk Again Project consortium — Nicolelis and collaborators from Europe and Brazil — was organizing and publishing position papers about an exoskeleton-based BCI system for spinal cord injury; the project would gain public attention at the 2014 World Cup opening ceremony but was conceptually advanced significantly in this period. P300 speller BCI research continued at Wadsworth and collaborating sites, with several papers on user performance, training protocols, and optimal matrix presentation rates in active submission.
April–June. The Schwartz group’s monkey self-feeding experiments, conducted using a four-degree-of-freedom robot arm (shoulder, elbow, gripper), had been completed by mid-2007 and the Nature manuscript was likely in preparation. The Journal of Neural Engineering published multiple papers on Kalman filter decoding of motor cortex spiking for robotic arm control, cementing the Kalman filter as the standard algorithm for closed-loop intracortical prosthetics. The Berlin BCI group (Blankertz, Müller, and colleagues) published key results on training-free motor imagery classification using the BBCI paradigm, showing that with appropriate signal processing — and without long calibration sessions — naïve users could achieve above-chance cursor control in their first BCI session. This addressed a major criticism of motor imagery BCIs, which had typically required weeks of training.
July–September. Matthew Nagle died on July 24, 2007, from complications of septicemia at 27 years of age. His passing was noted with eulogies from Donoghue and Hochberg, who emphasized that his participation in the BrainGate trial had been entirely voluntary and had provided the scientific community with irreplaceable evidence about human motor cortex function after SCI. Nagle was described in the BCI community as a pioneer in the truest sense — a patient who had enrolled “not for personal benefit” but to advance science that might one day help others. The ASHA Leader review of BCI speller technology appeared in September, providing a clinical community overview of P300 BCI systems and noting BCI2000’s use at more than 80 research centers globally. Nicolelis published a preliminary description of the avatar-control experiments at Duke, and gave public presentations describing the concept of a “brain-machine-body interface” in which a patient’s cortical signals could control an exoskeleton that would relay sensory feedback back to the brain.
October–December. The Society for Neuroscience annual meeting in San Diego (November 2007) included major BCI sessions, including presentations by Schwartz on the forthcoming self-feeding monkey data, by Shenoy on further improvements to premotor cortex decoding at Stanford, and by the EEG BCI community on advances in adaptive and training-free systems. Cyberkinetics was in advanced discussions with academic partners about transferring the BrainGate IDE to an academically based investigational device framework — the process that would yield the BrainGate2 IDE application in 2008. Blackrock Microsystems, which would eventually take over manufacturing of the Utah Array (previously Bionic Technologies/Cyberkinetics’ NeuroPort), was advancing its own 510(k)-cleared neural recording system. End-of-year NIH funding levels for neural prosthetics research remained substantial, with NINDS, NIDCD, and DARPA all maintaining or expanding their BCI portfolio investments.
Trends
Cyberkinetics’ Retreat and the Transition to Academic Trials
The financial difficulties and eventual exit of Cyberkinetics from BrainGate clinical research in 2007–2008 illustrated a structural tension that would recur repeatedly in neurotechnology: the gap between early scientific success and the commercial and regulatory path to a marketable product. Cyberkinetics had succeeded in demonstrating the BrainGate concept in four human participants; it had not succeeded in generating the commercial revenue stream needed to sustain operations. The consequence — a transition from a company-sponsored IDE to an academically sponsored IDE under Hochberg and Donoghue at MGH/Brown — was ultimately scientifically constructive, resulting in a more rigorous, NIH-funded research program. But it also left existing participants in an uncertain status and delayed enrollment. The lesson about the fragility of early commercial neurotechnology ventures would be relevant again when companies like Synchron and Neuralink emerged in subsequent decades.
Longevity of Neural Recordings: S3’s 1000+ Day Signal
While the Nagle trial had highlighted signal degradation as a concern, the experience of participant S3 (brainstem stroke) provided critical evidence that durable recordings were possible in some individuals. Her data, which would eventually be published in the Journal of Neural Engineering in 2011 (Simeral et al.) showing reliable cursor control at day 1000, were being accumulated throughout 2007. The emerging picture was that recording longevity was highly variable across participants — a function of individual biology, implant placement, the brain region targeted, and possibly the participant’s underlying condition and the status of the blood-brain barrier at the implant site. This variability motivated the field to look beyond recording longevity toward adaptive decoding algorithms that could compensate for gradual signal changes, rather than assuming stable recording as a prerequisite.
The Schwartz Monkey Self-Feeding Paradigm: Toward Natural Prosthetic Arm Use
Andrew Schwartz’s experiments, reported in a May 2008 Nature paper but largely completed in 2007, demonstrated that monkeys could use a four-degree-of-freedom robotic arm to reach, grasp, and deliver food to their own mouths with fluid, natural-appearing movements — while their own arms were restrained. The animals came to treat the robotic arm as part of their own body, licking marshmallow residue from the robot gripper and using it to maneuver food inside their mouths. This was a qualitatively different level of demonstration from cursor control: it showed that the motor cortex could control multi-joint, natural movement of an artificial limb in an ecologically meaningful task. The population vector decoder used by Schwartz, refined over years, was translating the summed directional preferences of 50–100 simultaneously recorded motor cortex neurons into smooth three-dimensional trajectories — a system optimized for naturalistic movement rather than speed of information transfer.
EEG BCI: Training-Free Systems and SSVEP Growth
The non-invasive community responded to the practical demand for systems usable by patients with limited training time or cognitive resources by intensifying work on training-free or minimal-calibration approaches. The Berlin BCI group’s training-free motor imagery classifier, which used the subject’s first session data to initialize a regularized spatial filter without extensive baseline periods, was a significant practical advance. The SSVEP paradigm — in which the user attends to one of several visual stimuli flickering at distinct frequencies (e.g., 6 Hz, 7.5 Hz, 10 Hz, 12 Hz), each producing a characteristic steady-state oscillation in occipital EEG — required essentially no training, provided reliable frequency-specific responses, and could support multiple simultaneous target selections. SSVEP BCIs were attracting attention as a complement to P300 and motor imagery paradigms, particularly for users who retained reliable gaze control but could not sustain the focused attention required for P300 detection across long sessions.
Neuromodulation and the BCI Crossover: Adaptive DBS
The growing appreciation that DBS electrodes recorded meaningful LFPs, not just delivered stimulation, was creating a conceptual bridge between neuromodulation and BCI. Papers from Brown’s group at Oxford showed that beta-band LFP power in the STN inversely correlated with therapeutic efficacy of DBS — higher beta power was associated with worse motor symptoms, and therapeutic stimulation suppressed beta power. This suggested that a DBS system monitoring its own LFP output could, in principle, detect the need for stimulation and administer it only when required, reducing total stimulation dose, extending battery life, and potentially improving the side-effect profile. The concept of “adaptive” or “closed-loop” DBS was clearly articulated in the scientific literature by 2007, though the first human demonstrations would wait until 2013–2016. The broader point — that implanted neural hardware should both record and stimulate, forming a true bidirectional interface — was becoming a consensus design principle.
Suggested Titles
- Pioneer’s Passage: Matthew Nagle, Cyberkinetics’ Crisis, and the BCI Field’s Growing Pains
- From Commercial to Academic: The BrainGate Transition and What It Means for Neural Interfaces
- Marshmallows and Motor Cortex: Schwartz’s Monkey Arms Point Toward Natural Prosthetic Control
- Day 1000 and Counting: Long-Duration BCI Recordings Rewrite the Longevity Narrative
- The Bidirectional Turn: BCIs That Read and Write Begin to Define the Next Decade