Open Seizure DetectorThere are several other projects with similar goals to OpenSeizureDetector. I have tried to give a comparison of the ones I know about below.
I am happy to receive suggestions for others to add or corrections if I have got something wrong.
Please contact graham@openseizuredetector.org.uk if you think anything needs to be changed, or if you have a better reference for the performance of a device.
| Project | Hardware | Detection Method | Evidence | Fault Detection | Internet Dependency | Alerting | Cost (3yr) | Sensitivity (Tonic-Clonic Seizures) | False Alarm Rate | Study Context |
|---|---|---|---|---|---|---|---|---|---|---|
| Open Seizure Detector | PineTime / Garmin | Motion (frequency analysis and Machine Learning (CNN)), HR Thresholds | Open source | Yes (watch ↔ phone monitoring) | None | Local alarm, SMS, WiFi | £35–£153 (no subscription) | ~80% [6] | ~1-2 per day for night time use [6] | Real-world usage |
| Empatica EpiMonitor | EmbracePlus | ML (ACC + EDA) | Peer-reviewed [1] | Partial (cloud-dependent) | Required | SMS / call | ~£728 (≈£299 + £143/yr) | ~95% [1] | ~0.25/day [1] | Multicentre clinical study |
| EpiWatch | Apple Watch | Motion, HR ML (Isolation Forest and classifiers) [11] | Peer Reviewed Papers [11] | Unknown | Yes (for alerts) | Call, SMS or Email | £727 (£187 + £180/yr) | 98% [11] | 0.08 /day | Hospital EMU (not real world) |
| SeizAlarm | Apple Watch | Motion + HR thresholds [2] | Vendor | Unknown | Optional | Call, SMS, email | ~£592 (≈£187 + £135/yr) | Not published | Not published | – |
| SmartMonitor Inspyre | Apple / Samsung | Rhythmic motion detection [3] | Published studies [3] | Partial | Optional | App, SMS, call | ~£457 (≈£187 + £7.50/mo) | High (~100% small studies) [3] | Specificity >90% [3] | Small clinical studies |
| NightWatch+ | Armband + base | ACC + PPG + posture filtering [5] | Peer-reviewed [5] | Partial | None | Local alarm | ~£1,208 (no subscription) | ~86% median [5] | Varies; see [5] | Prospective home study |
| Brain Sentinel SPEAC | Arm EMG | sEMG muscle activity | Clinical / FDA-cleared | Unknown | Unknown | Caregiver alert | ~£800+ (subscription unclear) | 100% agreement [4] | ~1.4/day [4] | Multicentre clinical |
| EpiHunter | EEG headband | EEG-based detection | Vendor + clinical | Unknown | Required (app/cloud) | App alerts | ~£2,000+ (≈£59/mo subscription) | >90% (absence seizures) [7] | Not reported | Clinical + real-world |
| Owlet | Custom Sock Sensor | PPG Pulse/O2 Sat. Thresholds | Unknown | Unknown | Optional [8] | Local Alarm, App Alerts [8] | £300 (no subscription) | Unknown | Unknown | Unknown |
| Emfit | Bed Sensor | Movement | Peer Reviewed Study [9] | Unknown | None | Local Alarm | £450 (no subscription) | 67% [9] | 0.007 per day (very low) | Epilepsy Monitoring Unit [9] |
| SAMI | Camera Based | Movement | Unknown | Unknown | None | Wifi (dedicated network) | £1005 (assuming VAT exempt). No subscription | Unknown, but good user reports [10] | Unknwon | Unknown |
References
[1] Empatica seizure detection study (Epilepsia) 1
[3] SmartMonitor study summary
[5] NightWatch study (Neurology)
[6] OpenSeizureDetector ML Algorithm performance using user-supplied data. Uses real-world data.
[7] EpiHunter performance summary
[10] SAMI3 Information, ratings and feedback
[11] EpiWatch - First Paper, EpiWatch - Second Paper
Notes:
- Fault Detection indicates whether the system detects device/link/alert failures to warn the user of problems rather than failing silently if it is not working.
- Internet Dependency distinguishes internet requirements for detection vs alerting purposes.
- Performance metrics are not directly comparable due to differing study designs and populations, but I have tried to quote published data where available.
- The table below is quite wide - there is a scroll bar at the bottom of the table to allow you to view the additional columns.
Key Observations
-
Subscription models dominate commercial offerings, often exceeding hardware cost over time.
-
Fault detection remains poorly specified across nearly all commercial systems.
-
Internet dependence is often understated Internet connectivity is required for many of the devices to deliver alerts - this could be an issue if it is to be used in a remote area.
Important Note on Comparisons
Even where performance numbers are available, they are not directly comparable due to differences in:
- study design
- patient population
- seizure definitions
- validation environments
We have tried to include available figures to give an indication of the likely performance comparison.
Projects in Development
| Project | Hardware | Notes |
|---|---|---|
| Ellow | Custom Hardware (PPG sensor) | Prototype stage. |
Other Projects that are No Longer Active
| Project | Hardware | Notes |
|---|---|---|
| Neutun | Pebble Watch | App no longer available, and company not active. (acquired by doc.ai?) |
| Brio | Heart Rate Sensor & Alarm | No Longer Available |
| Mellow | Apple Watch | App no longer available, and company not active. |
| PebbleSeizureDetect | Pebble Watch | Not actively maintained |