Can a single thought turn on a brain switch?

Imagine controlling a device with just your thoughts, without moving a finger. That’s what Neurodetector does, using electroencephalography (EEG) and event‑related potentials (ERP) to assess cognitive abilities. ERP are electrical spikes the brain produces when we pay attention; here they become a virtual “switch.”\n\nIn the lab, 40 healthy adults, mostly under 60, were asked to complete three tasks of increasing difficulty using a single mental button. The outcome? Everyone performed above chance, proving the brain can indeed command the device. Moreover, success rates dropped as tasks got harder, showing performance depends on each person’s cognitive capacity.\n\nThe researchers compared two ways to decode ERP: the classic peak‑based method and a new pattern‑matching approach. The latter delivered higher accuracy and was more sensitive to individual differences and task complexity.\n\nWhy does this matter? Because the EEG switch requires no physical movement, it opens the door to cognitive testing for people with motor limitations, such as the elderly or those with physical disabilities. Although the study involved only healthy subjects, the team aims to test the system on patients with mild cognitive impairment and dementia. Neurodetector could become an accessible, non‑invasive tool for monitoring mental health in the future.

Event‑related potentials (ERP) are time‑locked neural responses that reflect cognitive processes such as attention, working memory, and stimulus evaluation. In Neurodetector protocols, ERPs are evoked by brief visual or auditory cues while participants must recognize a target and, in response, trigger the EEG switch's state change.\n\nThe traditional decoding method relies on identifying specific peaks, for example the P300, which appears roughly 300 ms after the target stimulus. However, the latency and amplitude of these peaks vary widely across individuals and vigilance states, limiting detection robustness. The pattern‑matching introduced in the study employs supervised machine‑learning algorithms: raw EEG recordings are segmented into temporal windows, normalized, and then compared with a template set built from training data. Techniques such as cross‑correlation and convolutional neural networks (CNN) capture not only the main peak but the entire ERP spectral profile, including N200, P300, and late‑positive complex components.\n\nResults show pattern‑matching raises classification accuracy by 12‑15 % over peak‑based approaches, especially in more demanding tasks where cognitive variability is pronounced. Moreover, sensitivity to individual differences suggests the model can adapt to subjects with slower brain rhythms or high background noise, typical of older populations.\n\nClinically, this flexibility is crucial for applications to patients with mild cognitive impairment (MCI) or dementia, where ERPs may be attenuated or delayed. A pattern‑matching‑based interface could thus provide more reliable biomarkers for tracking disease progression or therapeutic efficacy.\n\nFuture research directions include integrating multimodal signals (EEG‑fNIRS, residual EMG) to boost command robustness, optimizing online learning algorithms to adapt in real time to cognitive state fluctuations, and conducting longitudinal studies on MCI cohorts to validate the EEG switch's predictive power for early dementia diagnosis.