Performance Summary

The following chart summarizes the performance of various seizure prediction technologies, including those discussed in this article. The x-axis represents the number of patients in the validation cohort, while the y-axis indicates the area under the curve (AUC) for each technology. The chart highlights the performance of different devices and algorithms, with blue dots representing forecasting technologies (e.g., Seer, Epiminder) and green dots representing real-time detection devices (e.g., NeuroPace, Empatica, UNEEG).

Blue dots represent forecasting technologies (e.g., Seer, Epiminder).

Green dots represent real-time detection devices (e.g., NeuroPace, Empatica, UNEEG).

The dashed vertical line marks the AUC 0.75 threshold — often considered a minimum bar for clinical prediction tools.

The gray shaded zone shows the high clinical utility region (e.g., FDA-approved, RCT-validated solutions).

Comparative Table of Seizure Prediction Technologies

The following table summarizes the key features and limitations of various seizure prediction technologies, including those discussed in this article. It provides a quick reference for understanding the differences between these technologies and their respective validation methods.

Notes:

• NeuroVista: An early intracranial EEG system that demonstrated the feasibility of seizure forecasting but was not commercialized.
• Seer Medical: Offers a wearable device paired with a mobile app for seizure forecasting; however, clinical outcome data is limited.
• Epiminder: Utilizes a subscalp EEG system showing high retrospective forecasting accuracy; prospective trials are pending.
• NeuroPace RNS: An FDA-approved intracranial device that provides real-time seizure detection and responsive neurostimulation, significantly reducing seizure frequency.
• Empatica Embrace: A wrist-worn device that detects generalized tonic-clonic seizures in real-time, alerting caregivers promptly.
• UNEEG SubQ: A subscalp EEG system designed for ultra-long-term monitoring, offering high sensitivity and specificity in seizure detection; currently used for diagnostics.

I. NeuroVista (2007–2014): Proof of Concept, But Not of Practice

What They Tried:

NeuroVista developed the first implantable seizure advisory system tested in humans. Their system included subdural electrodes connected to a chest-worn telemetry unit, transmitting real-time EEG data to an external device that displayed seizure risk via a color-coded light (blue = low, red = high).

Scientific Claims:

• Seizure prediction is possible using intracranial EEG in select patients [1].
• Personalized, within-subject algorithms could deliver warnings with >65% sensitivity.
• Seizure risk states (high/low) could be reliably distinguished above chance.

Validation:

• A 15-patient trial (2010–2013) with continuous recording [1].
• Subject-specific algorithms trained on initial data; tested prospectively within each patient.
• 11 out of 15 participants achieved better-than-chance prediction.
• Red warnings captured 56–100% of seizures in those responders.

Critical Flaws:

• One-third of participants did not benefit from predictions.
• High variability in algorithm success — some patients had no discernible predictive patterns.
• No clear quality-of-life improvement from using the device.
• Device-related complications (infections, hardware migration).
• Trial ended without commercialization; company shut down by ~2014.
• Data reuse in later studies may not account for limitations of original design.

II. Seer Medical (2017–2025): Big Data, Big Claims, Bigger Collapse

What They Tried:

Seer Medical built a platform combining wearable sensors, seizure diaries, and mobile apps to deliver personalized seizure forecasts based on circadian and multiday cycles. Their flagship product, the Seer Home EEG-ECG system, offered at-home brain monitoring over multiple days.

Scientific Claims:

• Seizure risk is cyclical in most people — based on circadian and multiday rhythms [2].
• These cycles can be detected from self-reported seizure times and wearable data [3].
• Machine learning models can generate personalized risk forecasts (AUC ~0.73–0.77).
• Forecasts empower patients to adjust behavior and avoid danger [4].

Validation:

• Retrospective studies using large seizure diary datasets (>100,000 seizures) [2].
• 2023 pilot study with 13 patients using wearable devices [4].
• Prospective validation in 6 participants: 4 showed significant above-chance forecasts.
• Use of ROC AUC and time-in-risk metrics; forecasts delivered via smartphone app.
• Prospective validation, though limited, was essential to moving beyond earlier retrospective claims.

Collapse:

• August 2024: FDA class II recall of Seer Home device due to EMC compliance failures [5].
• Ceased monitoring operations; closed clinics.
• Early 2025: Entered voluntary administration [6].
• Cadwell Industries acquired assets in distressed sale [6].

Critical Flaws:

• Publicly deployed forecasting tools before rigorous clinical trials.
• Forecasting often relied on patient self-reports, which are known to be unreliable [7].
• Limited clinical benefit data: no evidence that app usage improved outcomes.
• Regulatory oversight issues culminated in recall.
• Collapse wiped out ~\$30 million in venture and public funding.
• Cadwell likely purchased IP and data assets at negligible cost.

III. Epiminder (2018–Present): Next in Line?

What They're Trying:

Epiminder is developing the Minder™, a minimally invasive sub-scalp EEG system. It records 24/7 brain activity via a small electrode placed under the scalp and transmits data to an external behind-the-ear processor and smartphone app.

Scientific Claims:

• Continuous sub-scalp EEG can capture high-quality brain signals over months [8].
• EEG biomarkers (e.g., spike rate cycles) enable seizure risk forecasts.
• Retrospective forecasts in early patients achieved AUC up to 0.88.
• Device is safe, well-tolerated, and suitable for long-term use [8].

Validation:

• Phase I trial with 5 patients monitored for up to 12 months [8].
• Direct comparison with scalp EEG showed high concordance.
• Retrospective forecasting showed 83% of seizures occurred during "high-risk" periods occupying ~26% of total time.
• System performance promising but based on very small cohort.

Critical Risks:

• Forecasting performance shown only retrospectively; no prospective alerts tested yet.
• Sample size too small for generalization; trial ongoing.
• Implant, while less invasive than NeuroVista, still requires surgery.
• Commercial deployment not yet validated in regulatory trials.
• Shares scientific and institutional lineage with earlier efforts, raising concerns of history repeating.

The Bigger Pattern: When Medical AI Fails, It Fails Quietly

• NeuroVista's failure was a quiet end to a promising start, with no public accountability.
• Seer Medical's collapse left patients and investors in the dark, with no clear path forward.
• Epiminder's early promise is overshadowed by the failures of its predecessors.
• NeuroVista's data was reused in multiple studies, but the original flaws were never addressed.
• Medical prediction and detection models misfire and misdiagnose patients, creating unforeseen risks and questioning the reliability of AI-driven diagnostics.

And yet, the seizure prediction ecosystem has carried forward the same flawed assumptions, data, and optimism from one company to the next. The NeuroVista dataset seeded over 30 academic papers. Seer scaled those assumptions into consumer-facing tech. Epiminder is poised to follow a similar path — this time with better technology, but still without large-scale, prospective proof.

What Needs to Change

• Subject-independent, prospective validation must become the norm.
• Clinical trials demonstrating benefit before deployment.
• Full transparency of algorithm performance, including false positives and non-responders.
• Independent oversight in high-risk AI healthcare systems.
• Stop confusing exploratory modeling with medical-grade evidence.

Final Word

When a seizure-detection model misfires, it risks lives. When an LLM misfires, it sparks a Twitter thread. Yet somehow, we scrutinize the LLM more harshly — maybe because we understand language better than EEG?

The seizure prediction field has a long way to go before it can deliver on its promises. The failures of NeuroVista and Seer Medical should serve as cautionary tales, reminding us that the road to innovation is fraught with challenges. The Epiminder team is working hard to build a better future for patients with epilepsy, but they must learn from the mistakes of their predecessors. The future of seizure prediction is not just about technology; it's about understanding the complexities of the human brain and the ethical implications of our work.

The future of seizure prediction is bright, but it needs to be built on a foundation of rigorous science, ethical responsibility, and patient safety. We owe it to the patients who have suffered from the failures of the past to ensure that the next generation of seizure prediction technology is built on a foundation of rigorous science, ethical responsibility, and patient safety. Let's not let the next generation of seizure prediction tech repeat the mistakes of the past.

And let's remember: telling the truth about failure isn't cruel — it's care. Especially when that truth is offered with the intention to build something better. After $200 million in sunk ambition, can we really fear one more hard truth?

References

1. Cook, M. J. et al. (2013). Prediction of seizure likelihood with a long-term, implanted seizure advisory system in patients with drug-resistant epilepsy: a first-in-man study. The Lancet Neurology, 12(6), 563–571.
2. Baud, M. O. et al. (2018). Multi-day rhythms modulate seizure risk in epilepsy. Nature Communications, 9(1), 88.
3. Stirling, R. E. et al. (2021). Circadian and multiday seizure cycles are the norm in epilepsy. Brain, 144(9), 2696–2708.
4. Karoly, P. J. et al. (2023). Prospective validation of seizure forecasting with wearable and diary data. npj Digital Medicine, 6(1), 88.
5. FDA. (2024). Seer Medical Pty Ltd Seer Home Recall. Class 2 Device Recall Seer Home System
6. Business News Australia. (2025). Seer Medical enters administration, Cadwell Industries steps in. https://www.businessnewsaustralia.com
7. Haut, S. R. et al. (2002). Seizure self-report accuracy and reliability. Epilepsia, 43(5), 609–614.
8. Dumanis, S. B. et al. (2021). Chronic sub-scalp EEG monitoring and seizure forecasting in epilepsy. Epilepsia Open, 6(1), 1–10.

Disclaimer: This blog post is for informational purposes only and does not constitute financial, legal, or medical advice. The information provided is based on publicly available sources and should not be considered a substitute for professional advice. BioniChaos is not responsible for any actions taken based on the information provided in this blog post. ChatGPT was used to write this blog post. For further reading, check out the detailed OpenAI analysis in deepresearch.txt. Make sure to check any information you find here with the original sources.