# Is Tooth-Click Earable Control Ready for the Neurosurgical OR?

A new preprint from a team including researchers Jonas Hummel, Maximilian Burzer, Clara Sayffaerth, Valeria Zitz, Amir El Rahal, Michael Küttner, Tobias Röddiger, Jürgen Beck, and Michael Beigl proposes a hands-free earable control system for neurosurgical operating rooms — and the classification numbers are hard to ignore. Using the **OpenEarable 2.0** platform with tooth-click input from 12 participants, their real-time recognition pipeline achieved a **median macro F1-score of 98.6%** under leave-one-subject-out cross-validation. A follow-up evaluation with **20 neurosurgeons** during a simulated resection task found that participants reported few attentional disruptions and rated the system favorably for workflow integration and perceived safety. Autonomous microscope control was rated significantly higher with the earable system than with the Delegation condition (assistant-mediated control), though Delegation enabled faster task completion when a continuous assistant was available. Workload, usability, and task error rates showed no significant differences between conditions.

This is a small feasibility study — not a clinical trial — and the tooth-click input mechanism is non-neural. But the research sits at a direct interface with intraoperative [brain-computer interface](https://bciintel.com/glossary/brain-computer-interface) workflows and raises important questions about the future of hands-free intraoperative control platforms.

---

## What NeuroClick Actually Does

The core problem NeuroClick addresses is mundane but operationally significant: neurosurgeons operating under sterile field constraints cannot touch keyboards, touchscreens, or standard input devices. Current workarounds — voice commands, foot pedals, or delegation to a scrub tech or assistant — each carry tradeoffs in latency, noise sensitivity, workflow disruption, or dependence on staff availability.

The NeuroClick system routes tooth-click acoustic and inertial signals from an in-ear OpenEarable 2.0 device through a real-time classification pipeline. Tooth clicks — jaw-generated sounds and vibrations — are mechanically distinct, intentional, and producible without hand or voice involvement, making them plausible input candidates in high-sterility environments.

The design was grounded in formative OR observations and interviews with **10 domain experts** before any classifier was trained — a methodological choice that meaningfully differentiates this from purely lab-bench signal-processing work. The researchers observed actual neurosurgical workflow to identify the constraint points before proposing a solution.

The classifier's **98.6% median macro F1-score** under leave-one-subject-out cross-validation across 12 participants is a strong classification result, though it should be contextualized carefully: cross-validation performance in controlled signal acquisition does not automatically translate to real-OR robustness under acoustic noise, movement artifact, and the cognitive load of live surgery.

---

## The 20-Surgeon Evaluation: What It Shows and What It Doesn't

The simulated resection task evaluation with 20 neurosurgeons is the study's most clinically relevant data. Key findings from the source:

- **Autonomous microscope control** was rated significantly higher with Earable than with the Delegation condition
- **Delegation enabled faster task completion** when a continuous assistant was available
- **Workload, usability, and task errors** showed no significant differences between conditions
- **Preferences depended on** training level, system reliability, context, and assistant availability

The last point deserves emphasis. The system is not universally preferred — it's contextually preferred. Surgeons with more experience, operating in contexts where assistant availability is unreliable or where independence is operationally valued, rated it more favorably. This is realistic and honest characterization of where an early-stage intraoperative input modality actually sits.

The study was conducted in a simulated OR, not a live surgical environment. Simulated resection tasks approximate real workflow but cannot reproduce the full noise floor, cognitive loading, and unpredictable interruption patterns of actual neurosurgery. Any path to clinical deployment would require live-OR validation.

---

## Why This Is Relevant to the BCI Industry

NeuroClick is not a [brain-computer interface](https://bciintel.com/glossary/brain-computer-interface) in the strict sense — no neural signal is decoded. But its relevance to the BCI field is direct and multidimensional.

**First, intraoperative control is a shared problem.** Neurosurgeons performing procedures that involve intracortical mapping, ECoG strip placement, or intraoperative neural stimulation already interact with complex digital systems — imaging stacks, neuronavigation platforms, stimulator controls — under the same sterile-field constraints NeuroClick addresses. A validated hands-free input modality integrates directly into that workflow.

**Second, this demonstrates the earable platform's signal fidelity.** OpenEarable 2.0 capturing mechanically distinct tooth-click patterns with sufficient resolution for 98.6% F1 classification suggests the in-ear sensing modality has headroom for more complex physiological signal capture — including EEG-adjacent signals relevant to passive BCI monitoring of surgeon cognitive state. This connects to [affective BCI](https://bciintel.com/glossary/affective-bci) research examining real-time operator workload.

**Third, the regulatory pathway for intraoperative assistive input is cleaner than for implanted devices.** A non-invasive OR control interface bypasses the IDE, PMA, and De Novo complexity facing companies like [Neuralink Corp](https://bciintel.com/companies/neuralink), [Synchron](https://bciintel.com/companies/synchron), and [Precision Neuroscience](https://bciintel.com/companies/precision-neuroscience). If NeuroClick's performance holds in live-OR conditions, a 510(k) pathway is plausible. That makes clinical translation timeline potentially shorter than most BCI hardware in the pipeline today.

---

## Skeptical Read: The Gap Between Simulation and Sterilized Reality

The honest assessment is that **98.6% F1 in a controlled classification experiment is not the same as 98.6% reliability in a live craniotomy**. Several real-OR factors would stress-test the system:

- **Acoustic noise**: Bone saws, suction, electrocautery, and OR communication all generate artifact that may overlap with tooth-click frequency signatures
- **Head position variability**: Neurosurgeons work in fixed but sometimes extreme head positions relative to the microscope; earable fit and signal quality may vary
- **Training burden**: The study notes that preferences depended on training level, implying a learning curve that adds adoption friction in already time-pressured surgical teams
- **False positive consequences**: A misclassified tooth click that inadvertently advances a microscope focus plane mid-dissection has patient safety implications that are qualitatively different from a misclassified input in a consumer device

None of these are reasons to dismiss the work — they're the exact questions a follow-on live-OR trial would need to answer. The authors are transparent that this is a direct, hands-free option "for selected functions alongside established workflows," not a wholesale replacement for existing input modalities.

---

## Industry Trajectory and What Comes Next

The NeuroClick preprint is an arxiv submission (arXiv:2609.12910v1) as of publication date and has not yet undergone peer review. The appropriate weight to give it is: technically credible early-stage feasibility data that warrants follow-on investigation, not validated clinical evidence.

For the BCI and neurotech industry, the directional signal is that **intraoperative human-machine interface** is an underserved problem that non-neural sensing modalities may address faster than implanted BCIs. The convergence of high-accuracy in-ear sensing, low-latency classification pipelines, and genuine OR workflow research (the formative expert interviews are notable for a HCI paper) suggests a maturing methodology.

Companies building intraoperative neuronavigation and stimulation platforms — and the surgical robotics sector broadly — should monitor this work. If live-OR validation follows and maintains even 95%+ accuracy under realistic noise conditions, the path to OR adoption is shorter than most neural interface clinical translation timelines by several years.

---

## Key Takeaways

- **98.6% median macro F1-score** for tooth-click classification using OpenEarable 2.0 across 12 participants under leave-one-subject-out cross-validation
- **20-neurosurgeon simulated OR evaluation** found significantly higher autonomy ratings for earable vs. delegation, but delegation was faster when assistants were continuously available
- Workload, usability, and task errors showed **no significant differences** between earable and delegation conditions
- Design was grounded in formative OR observations and interviews with **10 domain experts** — not purely a lab exercise
- This is a **preprint feasibility study**, not a peer-reviewed clinical trial; live-OR validation is the critical next step
- Regulatory path for a non-invasive intraoperative input device is substantially simpler than for implanted BCI hardware
- The earable sensing modality has potential headroom for passive physiological monitoring beyond tooth clicks

---

## Frequently Asked Questions

**What is NeuroClick?**
NeuroClick is a hands-free control system for neurosurgical operating rooms that uses tooth-click input detected by an in-ear OpenEarable 2.0 device. A real-time classifier recognizes intentional tooth clicks to control OR equipment — such as the surgical microscope — without requiring hand or voice input.

**What accuracy did the NeuroClick classifier achieve?**
The classifier achieved a median macro F1-score of 98.6% under leave-one-subject-out cross-validation across 12 participants. This was measured in a controlled signal acquisition setting, not a live surgical environment.

**Is NeuroClick a brain-computer interface?**
No — NeuroClick does not decode neural signals. It detects jaw-generated mechanical events via an in-ear sensor. However, it addresses the same intraoperative workflow constraints relevant to surgical procedures involving neural interfaces, and the earable sensing platform has potential extension to passive neural or physiological monitoring.

**Was NeuroClick tested with real neurosurgeons?**
Yes. Twenty neurosurgeons evaluated the system during a simulated resection task in a neurosurgical OR. This is a small feasibility evaluation, not a randomized controlled trial. Results from simulated tasks may not fully generalize to live surgical conditions.

**What are the main limitations of this research?**
The study was conducted in a simulated OR environment, not during live surgery. The 12-participant classifier training set is small. Real-OR acoustic noise, head positioning variability, and the consequences of false-positive inputs in a live craniotomy remain untested. The preprint has not yet undergone peer review.

**What would need to happen before NeuroClick could be used in actual surgery?**
Live-OR validation under realistic noise and workflow conditions, peer-reviewed publication, and likely FDA 510(k) clearance as a surgical assistive device. The non-invasive, non-neural nature of the system means the regulatory pathway is simpler than for implanted BCI hardware, but OR safety standards for input devices remain stringent.

---

*This article is based on a preprint feasibility study (arXiv:2609.12910v1) and does not constitute medical advice. Results reflect a small simulated-OR evaluation and should not be interpreted as evidence of clinical efficacy or safety for use in live neurosurgical procedures.*