Neural NoiseBCI Annual Review — 2010
1 January–31 December 2010
Introduction
By 2010 the Brain-Computer Interface field had stabilized into a recognizable institutional shape: a federally funded academic core organized around BrainGate2 at Brown/MGH, a growing European BNCI (Brain/Neural Computer Interaction) consortium, a commercially maturing cochlear implant industry, and an advancing Argus II retinal prosthesis trial working toward regulatory approval. The year’s defining theme was the consolidation of methods — particularly the common spatial pattern (CSP) filter for EEG, the expanding utility of electrocorticography, and the emergence of hybrid BCI architectures — rather than any single dramatic clinical milestone. The BCI Competition IV, whose results had been formally announced at the NIPS 2008 workshop, continued to generate derivative publications as laboratories refined decoding algorithms on its benchmark datasets, and the review article by Blankertz, Tangermann, and colleagues was in preparation (it would appear in Frontiers in Neuroscience in 2012), constituting the field’s most systematic accounting of non-invasive BCI algorithm performance to date.
BrainGate2 (NCT00912041) enrolled its first participant under the new academic protocol in 2009 and was actively accruing data from participant T1, a person with ALS, in 2010. The multi-site infrastructure — with active research sites at Massachusetts General Hospital, the Providence VA Medical Center, and Stanford University — was generating intracortical recording sessions with the Blackrock Microsystems NeuroPort 96-channel Utah array, the instrument that had replaced the Cyberkinetics recording system. The basic scientific agenda for the year included high-density characterization of human motor cortex spiking during attempted arm movement, assessment of signal longevity across months, and iterative refinement of Kalman filter decoders for continuous cursor control. Published work from the trial participants enrolled in the original 2004–2008 phase continued to appear, including analyses by Ajiboye, Hochberg, Donoghue, and Kirsch at EMBC 2010 characterizing imagined single-joint movements in high tetraplegia.
In ECoG, the field was expanding rapidly beyond simple motor paradigms. Gerwin Schalk’s group published two-dimensional cursor control results; Eric Leuthardt’s laboratory at Washington University in St. Louis was working on paradigms that engaged speech and language networks on the cortical surface, generating preliminary results on phoneme articulation decoding that would be fully reported in 2011. Zhuang, Truccolo, Vargas-Irwin, and Donoghue published in IEEE Transactions on Biomedical Engineering in 2010 on decoding three-dimensional reach and grasp kinematics from high-frequency local field potentials in motor cortex, extending the decoded parameter space from cursor velocity to continuous limb trajectories. On the clinical neurostimulation side, a JAMA Neurology expert consensus workshop published its comparison of subthalamic nucleus (STN) versus globus pallidus interna (GPi) deep brain stimulation targets for Parkinson’s disease, sharpening the clinical decision framework that would inform subsequent head-to-head trials.
Timelines
January–March. The early months of 2010 were marked by continued algorithm development activity in the non-invasive BCI community. The CSP filter, originally developed by Ramoser, Müller-Gerking, and Pfurtscheller in the late 1990s, had by this point become the consensus preprocessing step for motor imagery EEG decoding, and a wave of variants — regularized CSP, common spatio-spectral patterns, CSP combined with spectral filters — were being benchmarked against the BCI Competition IV datasets. Pfurtscheller’s group at Graz University published a conceptual overview of hybrid BCIs, defining the architectural space of systems that combined two or more BCI subsystems (motor imagery plus P300; EEG plus EMG; EEG plus eye tracking) to improve overall information transfer rate and broadened usability beyond the subset of subjects capable of reliable motor imagery control. The DBS community published the consensus expert workshop article in JAMA Neurology in early 2010, synthesizing the evidence on STN versus GPi targets for PD and setting the stage for the Veterans Affairs cooperative study results that would appear in 2011.
April–June. Spring 2010 saw BrainGate2 enrollment expanding. Participant T1, an individual with ALS, was actively enrolled at the Providence VA site, and data collection sessions characterized high-density spiking from a 96-channel Utah array implanted in left motor cortex during attempted and imagined upper-limb movements. Ajiboye and colleagues presented at the IEEE EMBC 2010 conference in Buenos Aires decoding results for imagined single-joint movements in a tetraplegic participant, demonstrating that discriminative neural signals for individual joints (shoulder flexion, elbow extension, wrist rotation) could be extracted even from a cortex that had been deprived of proprioceptive feedback for years. The cochlear implant market absorbed Cochlear Americas’ announcement of an upgrade program for Nucleus Freedom recipients to the Nucleus 5 platform, a backward-compatibility commitment that reinforced loyalty among existing implant recipients and set an industry precedent. The European BNCI Horizon 2020 project was actively framing six application themes for brain/neural computer interaction, mapping the regulatory and scientific needs of the field at a European level.
July–September. The summer of 2010 was productive for ECoG and LFP decoding research. Zhuang, Truccolo, Vargas-Irwin, and Donoghue’s paper in IEEE Transactions on Biomedical Engineering appeared, reporting that high-frequency (200–400 Hz) local field potential power in motor cortex could decode three-dimensional reach and grasp kinematics with accuracy approaching that of spike-sorted single units, but without requiring the identification of individual neurons. This result was significant for translational BCI engineering: if LFPs rather than spikes could support accurate decoding, chronic recording systems might be less vulnerable to the single-unit signal degradation known to occur over months of implantation. Separately, Schalk and colleagues continued ECoG 2D cursor control work that demonstrated accuracy rates competitive with single-unit BMI systems in epilepsy patients undergoing presurgical monitoring, reinforcing ECoG’s case as an intermediate-invasiveness platform. Second Sight Medical Products continued accruing Argus II trial data from thirty enrolled subjects, with longer follow-up durations now providing multi-year safety data that supported the CE Mark application being prepared for European submission.
October–December. The final quarter of 2010 saw significant review and synthesis activity. The BCI field’s end-of-year literature was increasingly focused on the gap between laboratory demonstrations and clinical utility. Allison and colleagues published analyses of BCI illiteracy — the 15–30% of able-bodied subjects who could not achieve reliable motor imagery BCI control — identifying neurophysiological and psychological predictors. The Brunner ECoG speller, a grid-based ECoG P300 speller demonstrated in epilepsy patients achieving approximately 17 characters per minute, was circulating in conference proceedings and demonstrated that ECoG could support communication BCIs with substantially higher information transfer rates than scalp EEG P300 systems. DBS device utilization continued its steady growth, with Medtronic reporting that more than 80,000 patients worldwide had received implantable pulse generators for movement disorders, pain, and psychiatric indications, making DBS the second most numerically deployed implanted neural interface after cochlear implants. Blackrock Microsystems consolidated its position as the dominant supplier of research-grade Utah arrays, supporting BrainGate2 sessions and growing numbers of academic electrode physiology laboratories internationally.
Trends
CSP and the Maturation of Non-Invasive Decoding
The common spatial pattern filter emerged in 2010 as the uncontested core preprocessing method for motor imagery EEG BCIs. Its dominance in the BCI Competition IV results had been anticipated, but by 2010 its role was institutionalized: algorithms submitted to competitive benchmarks and reviewed in journals were routinely compared against CSP baselines, its mathematical properties were well-understood in terms of its relationship to Fisher’s linear discriminant and generalized eigenvalue decomposition, and its limitations — sensitivity to outliers, poor performance on subjects with atypical motor imagery EEG — were mapped. The consequence was a bifurcation of non-invasive BCI research: one stream continued refining CSP variants (sparse CSP, subject-independent CSP, spatio-temporal extensions), while a second stream acknowledged that motor imagery remained unreliable for a substantial minority of users and turned toward P300 and SSVEP paradigms where signal generation was more externally driven and subject independence was higher.
ECoG as a Clinical Research Platform
By 2010, electrocorticography had been transformed from an incidental recording opportunity during epilepsy presurgical monitoring into a systematic research platform with dedicated protocol infrastructure at several centers. The appeal was structural: epilepsy patients undergoing temporary subdural grid implantation provided weeks of high-bandwidth cortical surface recordings in awake, cooperative subjects under conditions of patient safety imperative (i.e., no additional surgical risk beyond the clinically required procedure). The research access this provided — particularly to sensorimotor, language, and association cortex — was enabling ECoG work that was impossible in the scalp EEG laboratory and that did not require new invasive procedures. The Leuthardt and Schalk groups at WUSTL and Albany, and emerging groups at UCSF, were each developing paradigms covering motor, speech, and cognitive domains. The absence of an approved chronic ECoG device for BCI application (all recording was done with existing clinical grid electrodes) remained a technical and regulatory constraint, but the scientific momentum was undeniable.
Hybrid BCIs and the Usability Problem
Pfurtscheller’s hybrid BCI framework, elaborated in 2010, addressed a fundamental usability challenge: no single BCI control paradigm worked reliably for all users. Motor imagery BCIs failed the illiterate 15–30%; P300 BCIs required attention to a flickering stimulus and were exhausting over extended use; SSVEP BCIs required fixation and suffered in users with poor visual acuity. Hybrid architectures — combining two paradigms (e.g., SSVEP for menu selection and motor imagery for confirmation) or integrating BCI output with auxiliary physiological signals (EMG, EOG) — offered a practical path to higher reliability and broader usability without requiring any single paradigm to be universal. The hybrid BCI framework also connected with the emerging concept of passive BCIs, in which continuous physiological monitoring (EEG workload, attention) informed adaptive computer interfaces without the user explicitly generating command signals.
Deep Brain Stimulation: Clinical Consolidation
With over 80,000 implants worldwide by the end of 2010, deep brain stimulation had crossed from experimental therapy to mainstream neurological treatment for Parkinson’s disease, essential tremor, and dystonia. The 2010 JAMA Neurology consensus workshop publication on STN versus GPi targeting formalized the clinical decision framework: STN offered greater energy efficiency (lower stimulation amplitudes, longer battery life) and some data suggesting it could reduce dopaminergic medication requirements; GPi offered a more favorable neuropsychiatric profile and greater flexibility in stimulation programming without cognitive side effects. The VA cooperative study (Weaver et al., JAMA 2009) had provided the first randomized multicenter evidence confirming similar motor outcomes between the two targets but differing psychiatric profiles, and the 2010 expert synthesis incorporated these findings into practical programming and patient selection guidance. The field was beginning to debate the next frontier: responsive, closed-loop DBS that used neural biomarkers — in particular, the pathological beta oscillation in the basal ganglia LFP — to trigger stimulation only when needed.
Signal Longevity and the Chronic Recording Problem
By 2010, the most honest accounts of intracortical recording systems acknowledged a consistent pattern: signal quality — as measured by spike amplitude, unit yield per electrode, and decoder performance — declined over months of chronic implantation, with most arrays losing the majority of well-isolated single units within one to two years. The biological substrate for this degradation — a neuroinflammatory cascade triggered by micromotion of the rigid silicon shank, leading to microglial activation, astrocytic encapsulation, and neuronal retreat — was being characterized with increasing precision by Christine Kipke’s, William Shain’s, and Ravi Bellamkonda’s groups. Multiple approaches were in experimental development: softer, more compliant electrode materials; anti-inflammatory surface coatings; smaller electrode footprints; and the LFP-based decoding strategy demonstrated by Zhuang and colleagues, which reduced dependence on well-isolated single units. The BrainGate2 trial was itself generating the most systematic longitudinal human dataset, with participant S3’s array (implanted in 2005) now approaching its fifth year — a duration that would prove pivotal in the 2011 and 2012 publications.
Retinal Prosthetics Approach the Regulatory Threshold
The Argus II epiretinal prosthesis was, by the end of 2010, the most clinically advanced sensory restoration device after cochlear implants and arguably the most imminent candidate for commercial approval among visual prosthetics. With thirty enrolled subjects across multiple centers and two to three years of post-implant follow-up for early enrollees, the safety data were favorable and the functional vision outcomes — measured on standardized tests of square localization, direction of motion discrimination, and grating acuity — consistently exceeded device-off baselines. Second Sight was preparing its CE Mark application for Europe, having met the requirements of the CE regulatory process under the European Medical Devices Directive. Approval in 2011 would represent the first commercial retinal prosthesis in history, establishing a template for visual neural prosthetics regulation that the nascent cortical visual prosthetics field (Dobelle, phosphene stimulators) lacked.
Suggested Titles
- Methods in Consolidation: CSP, ECoG, and the Architecture of a Maturing Field
- The Hybrid Turn: When No Single BCI Paradigm Suffices
- Five Years in the Skull: Signal Longevity and the Chronic Recording Problem
- DBS at Scale: 80,000 Implants and the Beta Oscillation Frontier
- ECoG’s Moment: From Epilepsy Surgery to BCI Research Platform