How Epilepsy Patients Are Using Neurotech to Predict Seizures Before They Happen
From wrist sensors to brain implants, a wave of AI-powered devices is giving epilepsy patients what they've wanted most: a warning.
Imagine planning your day around the possibility that, at any moment, without a flicker of a warning, your brain might short-circuit. No countdown. No aura. Just the floor rushing up to meet your face. For the roughly 50 million people worldwide living with epilepsy, that’s not a thought experiment — it’s Tuesday. About 30 to 40 percent of them have drug-resistant epilepsy, meaning medication has already failed them and seizures keep coming regardless. The only thing missing, the thing patients and neurologists have wanted for decades, is a forecast. A weather warning for the brain. 🧠
That forecast is no longer purely hypothetical. A convergence of machine learning, implantable neurotechnology, and increasingly clever wearable sensors is making it possible to predict seizures before they happen — sometimes minutes in advance, sometimes hours. It’s not perfect. It doesn’t work for everyone. But the science has moved fast enough that real devices are shipping, real studies are reporting real numbers, and real patients are changing how they live their lives because of it.
Detection vs. forecasting: why the distinction actually matters
People tend to use “seizure detection” and “seizure prediction” as if they mean the same thing. They don’t, and the difference is enormous. 🔬
Seizure detection means a device notices a seizure is already happening and alerts a caregiver. That’s useful — it can speed up emergency response and reduce the time someone lies unconscious on a bathroom floor. But it doesn’t give the patient any agency. The event has already started.
Seizure forecasting is a different animal entirely. It means a device reads physiological signals — brain waves, heart rate, skin conductance, temperature — and tells you before anything goes wrong that your risk is elevated. A few minutes of warning, according to researchers at Mayo Clinic, can mean:
Calling a caregiver or family member
Sitting or lying down in a safe place
Avoiding a car, a swimming pool, or a busy street
Potentially triggering an automatic intervention like brain stimulation
That’s not a trivial difference. That’s the difference between being a passive victim and having some control over your own safety. Dr. Benjamin Brinkmann, a biomedical engineer at Mayo Clinic, has put it bluntly: the idea is simple — give people a warning. His team, working with both smartwatches and implanted devices, has correctly predicted approximately 75% of seizures with few false alarms in early studies. 🎯
The reason this is hard, though, is worth understanding. A seizure isn’t a discrete event that switches on suddenly. It’s the culmination of shifting electrical states in the brain, and those states can be nudged by sleep quality, stress, hormones, medication timing, and factors nobody has identified yet. Teaching a machine to read those signals reliably — across different patients, different seizure types, different lives — is genuinely difficult.
The wearables getting closest to the market
The most accessible tier of this technology lives on your wrist or behind your ear, and a few companies have gotten close enough to shipping that patients can actually buy something today. 💡
Empatica’s EpiMonitor, the successor to the widely recognized Embrace2, is the only FDA-cleared wearable for seizure detection currently available for purchase in the US. The system pairs the EmbracePlus medical smartwatch with a companion app, detects generalized tonic-clonic seizures via a smart algorithm with a 98% accuracy rate, and has up to seven days of battery life. It monitors electrodermal activity, temperature, accelerometry, and movement. Right now, it’s primarily a detection device. But Empatica is actively running a first-of-its-kind study to develop a full seizure forecasting algorithm using real-world data from EpiMonitor users across the US — which suggests the detection hardware is already collecting the data that forecasting models need.
Then there’s EPISERAS, a device from Spanish startup mjn-neuro that announced a European launch partnership with Neuraxpharm in late 2025. It’s an in-ear sensor, discreet as a hearing aid, that continuously records brain activity and fires an alert minutes before a seizure occurs. The company has been building this since 2014, has clinical evidence from multicenter studies across Spain, the UK, and Germany, and calls it the first digital health solution designed for real-time early detection in both ambulatory and home-care settings. The European launch is scheduled for 2026.
And then there’s Theta Neurotech, a startup out of the University of Chicago with a device that looks almost mundane: two adhesive sensors worn behind the ears. Under the hood, a machine-learning model detects subtle neurological shifts and can alert patients up to two hours before a seizure. Two hours. That would change daily life in fundamental ways.
Key differences between these wearable approaches:
EpiMonitor (Empatica): FDA-cleared, commercially available now, currently detection-focused, forecasting study underway
EPISERAS (mjn-neuro/Neuraxpharm): in-ear EEG, real-time risk prediction, European launch 2026
Theta Neurotech Patch: behind-the-ear EEG sensors, up to 2-hour advance warning, still in development
Brain Sentinel SPEAC: arm-worn sEMG monitor, the other FDA-cleared non-EEG option, detection-focused 🧬
Have you tried any epilepsy wearable, or know someone who has? The gap between clinical trial results and real-world usability is often the most telling data point of all — drop your experience in the comments.
The implant that talks back to your brain
Wearables are compelling for accessibility, but they face a stubborn limitation: they’re reading peripheral signals — skin conductance, movement, heart rate — and inferring what’s happening in the brain. Implants skip the inference step entirely. ⚡
NeuroPace’s RNS System is the most mature player in this space. It’s a closed-loop brain-computer interface: a neurostimulator implanted in the skull, with leads placed directly at the seizure focus. The device continuously monitors brain waves, recognizes each patient’s unique seizure-onset pattern, and responds with brief electrical pulses to interrupt the abnormal activity before a full seizure can develop. It’s personalized from day one, since every patient’s brain has its own electrical fingerprint.
The long-term data on the RNS System is genuinely striking. At the 2025 American Academy of Neurology meeting, NeuroPace presented three-year post-approval study data from 324 patients across 32 centers — the largest FDA-reviewed prospective neuromodulation trial ever conducted. The results showed:
82% median seizure reduction at three years
42% of patients seizure-free for six months or longer
62% median seizure reduction achieved as early as six months post-implant
Continued improvement over time, not a plateau
That last point matters. Most seizure medications provide their best response early and then level off. The RNS system keeps getting better as it learns its patient’s brain patterns over months and years. There’s something almost elegant about a device that improves with experience.
The tradeoff, of course, is surgery. The implant isn’t a good fit for everyone, and it requires patients to upload their brain data regularly via a remote monitor. It’s also currently FDA-approved only for adults 18 and older with focal-onset seizures, though NeuroPace has active research into Lennox-Gastaut Syndrome, a severe childhood epilepsy that has historically had very few treatment options. The company received a $9.3 million NIH BRAIN Initiative grant to fund that feasibility study.
Why your brain runs on a schedule (even if you don’t know it)
Here’s something that might reshape how you think about seizures: they aren’t random. Or rather, they’re less random than most patients believe. 📈
Research published in Clinical Epileptology and across multiple journals has shown that seizure patterns in many patients follow circadian rhythms — daily cycles tied to the sleep-wake cycle — and multidien rhythms, meaning multi-day patterns that can span weeks or even months. A 2024 review in Frontiers in Neurology notes these patterns are present in as many as 30% of people with drug-resistant epilepsy. Some patients have weekly seizure cycles; others monthly. Some have cycles that correlate with menstrual hormones. Some have cycles that nobody has explained yet.
This rhythmicity is why seizure forecasting is even theoretically possible with non-invasive wearable devices. If your brain runs a loosely predictable schedule, then a machine learning model fed weeks of your heart rate variability, skin conductance, temperature, and movement data can start to spot where you are in that cycle — even without reading your EEG directly. A 2021 study published in Scientific Reports from Mayo Clinic showed a wrist-worn LSTM neural network achieved an AUC-ROC of 0.80 for seizure forecasting, beating random prediction in 5 of 6 patients studied in real ambulatory settings, with concurrent confirmation from an implanted recording device.
The AI isn’t predicting a specific moment. It’s estimating risk windows — periods where probability is elevated enough to warrant caution. Think of it less like a car’s collision warning and more like a weather app saying 70% chance of rain. You don’t know exactly when it’ll rain. You bring an umbrella anyway.
Key signals wearable forecasting systems are currently learning to read:
Heart rate variability and cardiac rhythms synced with seizure cycles
Electrodermal activity (skin conductance), which reflects autonomic nervous system shifts
Body temperature and sleep quality patterns
Accelerometry for movement and rest cycles
Multiday rhythms identified from weeks of continuous data collection 🔬
The hard part is that these rhythms are deeply individual. A model trained on a population doesn’t translate cleanly to any single patient. The best systems are therefore personalized, trained on weeks of data from the specific patient before they start generating forecasts. That’s a slow, data-hungry process — but that’s exactly what devices like EpiMonitor and the NeuroPace RNS are designed to enable.
What’s still missing, and why this isn’t solved yet
Let’s be clear about something: this technology is real, and it is genuinely improving lives. But the honest picture is messier than the press releases suggest. 🧬
The clearest gap is coverage. Current forecasting and detection systems work best — sometimes only — for generalized tonic-clonic seizures, the convulsive type that’s visually dramatic and physiologically distinct. Focal aware seizures, absence seizures, and other subtler types are far harder to detect, let alone predict, because their physiological footprint in peripheral signals is much smaller. A 2025 scoping review from Weill Cornell Medicine in Qatar, published in JMIR, found that while AI-driven wearables show significant promise, the field still struggles with standardized validation methods, small study populations, and the challenge of deploying energy-efficient algorithms in real-world settings.
There’s also a false alarm problem. Every false alarm erodes trust. If a device alerts you five times and none of them result in a seizure, you start ignoring alerts on the sixth — the one that’s real. Getting sensitivity and specificity right at the same time, across diverse patients, outside of controlled hospital conditions, remains a stubborn engineering challenge.
And then there’s the question of what to do with a warning. If you get a 20-minute heads-up, what happens next? Sitting down is a good answer. But the bigger, more interesting answer is closed-loop intervention — where the warning itself triggers a treatment response, like a burst of medication delivered precisely, or targeted brain stimulation via an implanted device. That’s the direction Dr. Brinkmann’s work at Mayo Clinic is explicitly pointed. The idea of a seizure warning system that also prevents the seizure, automatically, without the patient having to do anything, is as compelling as it sounds.
Consider this: if a wearable could give you a reliable 30-minute window of elevated risk and also automatically notify emergency services or a caregiver, would you wear it every day? And what would it cost — financially, emotionally, in terms of the constant surveillance of your own body — to live with that kind of system forever? That question might be harder than the engineering. If you’re living with epilepsy, or close to someone who is, what would a reliable seizure forecast actually change for you? The comments are open.
For anyone wanting to understand how devices like these fit into the broader trajectory of brain-computer interfaces, NeurotechMag’s own look at five neurotech devices you can actually buy today and the signals that neurotech is reaching a tipping point are worth your time. The seizure prediction space isn’t a niche medical story — it’s one of the clearest proofs that neurotechnology has moved from lab curiosity to something that hands people real agency over their own biology.
The next step for the field isn’t more research showing the technology can work. It’s proving it works reliably enough, cheaply enough, and comfortably enough that the 90% of epilepsy patients who currently have no warning at all can finally have one. That number — 90% with no warning — is the number the whole field is trying to shrink.