Neuro Scan AI Solutions is a Finnish AI & robotics company building clinical-grade epilepsy diagnostics — pediatric-first, where the diagnostic gap is widest. The wearable EEG headset and mobile app are built; the detection engine is validated on the CHB-MIT pediatric dataset. Live hospital deployment is the next gate — ethics-committee reviews are in progress at four sites (two in Turkey, one in Helsinki, one in the United States).
Up to twenty-four electrodes positioned by the international 10–20 system. Bipolar montage at 256 Hz. Designed for ambulatory home use as well as in-clinic monitoring, with on-device buffering for connectivity loss and encrypted upload.
The NSAS reference device is a wireless, wearable headset designed for comfortable long-term wear at home and clinical-grade fidelity in the ward. Silver/silver-chloride electrodes, 10–20 international placement, paired bipolar montage. Compatible with existing clinical EEG hardware via standard data export. CHB-MIT validation uses the 22-channel bipolar standard.
Every NSAS prediction is auditable. No black-box deep nets where we can't show the responsible features. Random Forest + post-processing, with full provenance from electrode to alert.
22-channel scalp EEG streams from clinical hardware at 256Hz. Bipolar montage, real-time.
→The continuous signal is sliced into non-overlapping 5-second windows for downstream feature work.
→92 features computed per window — time, frequency, entropy, statistical. Top 32 selected per channel.
→Random Forest segment classifier. Auditable, feature-explainable — clinicians can trace any decision back to features.
→Post-processing groups consecutive segments into seizure events. Smart escalation to clinician via NSAS Mobile.
Reported with all continuous EEG data (not curated subsets). Cross-subject splits — no patient leakage between train and test.
Sensitivity is the harder metric: did the system catch the actual seizure? We report it openly alongside accuracy, on continuous data, with no patient leakage.
NSAS Mobile is built to stream the patient's home EEG, surface detected events in real time, and route alerts to the care team — without the bureaucratic dance of legacy patient portals. The app is ready; live patient streaming opens with the first clinical pilot.
Continuous 22-channel feed with on-device buffering when the network drops.
Push notification within seconds of detection — patient, family, and clinician.
Replay any event with the surrounding EEG context. Export for the neurologist.
One-tap share with the care team via GDPR-compliant secure channels.
CHB-MIT is a publicly available continuous scalp EEG dataset from Boston Children's Hospital, collected in collaboration with MIT. NSAS reports performance across the full continuous recordings — not curated subsets.
Continuous multi-channel scalp EEG from 24 pediatric patients with intractable seizures. Recorded after withdrawal of anti-epileptic medication to characterize seizures and assess candidacy for surgical intervention. The de-facto benchmark for any serious seizure detection system.
Many published baselines test on hand-picked records that contain seizures. NSAS reports on all data, including long stretches of non-seizure activity.
No patient appears in both train and test. Leave-one-out and 5-fold subject-wise — preventing the data-leakage that inflates benchmark numbers.
Output is seizure events, post-processed from segment classifications — the unit clinicians actually act on.
Cross-subject generalization, continuous-data evaluation, and event detection — the three things most prior CHB-MIT studies skip. NSAS' methodology builds directly on this framing: 92 features, top-32 selection, RF classification, event post-processing.
Read the paper →The Shoeb & Guttag paper that defined the patient-specific seizure-detection benchmark. 96% sensitivity on 173 seizures, 3-second median delay, 2 false detections / 24h. The reference point for any new approach on CHB-MIT.
Read the paper →The dataset itself — Boston Children's Hospital + MIT collaboration. 24 pediatric patients, 916 hours, 173 annotated seizures. Open access for academic research.
View the dataset →How we balance feature richness with deployment cost: time-domain, frequency-domain, entropy, and statistical features computed in 5-second windows, then ranked per-channel by mutual information.
Request the note →Oto (Meritxell). The source for the misdiagnosis figures we cite on this site: even at tertiary epilepsy centres under specialist care, misdiagnosis rates over 20% have been reported. The clinical motivation for everything NSAS does.
Read the paper →Christensen et al. The source for the per-patient cost figures on this site (~€15,240/year in Finland, ~€10,650/year in high-income countries). What the system spends today on a problem we believe is partially addressable.
Read the paper →Honest status, in one section. We have shipped the parts we control — model, app, hardware — and we are in the ethics-review queue at the four hospitals that will host our first live pilots.
CHB-MIT validation. 92 features, top 32 selected. Random Forest. Cross-subject splits.
Expo + React Native. Patient and clinician flows. Push alerts, event replay, GDPR exports.
22-channel scalp EEG, 256 Hz, bipolar montage. Compatible with standard clinical EEG interfaces.
Submissions open at 2 hospitals in Turkey, 1 in Helsinki, and 1 in the United States. Partners published on consent.
Scoped to EMU-equivalent settings. Real EEG, real clinicians, quarterly performance review on site data.
Class IIa software-as-medical-device pathway. ISO 13485 QMS in build. Submission depends on pilot data.
NSAS sits at the bridge between a dataset-validated algorithm and a CE-marked clinical product. Capital deployed at this stage funds the clinical-pilot operations at the four hospitals under ethics review, the regulatory submission package, and the small team that will deliver both.
We are deliberate about who we work with. If a healthcare-adjacent fund or strategic with neurology-domain depth resonates, we want to hear from you.
View the investor briefA 30-minute walkthrough with a neurologist, an engineer, and a regulatory lead. We bring the platform; you bring the questions. No patient EEG required to evaluate the workflow.