Month: February 2026

A new draft of the Version 5.0 release (V5.0.0g) is now available for testers (you need to ask for access to this at the moment, because it is not quite ready for ‘production’ use).

The changes in this draft are:

• Fixed an issue with the user interface not correctly showing ‘FAULT’ if the watch is out of range, and continuing to emit alarm sounds if it was in alarm when the watch went out of range.
• Fixed a problem with the app using a lot of data storage on the phone.
• Tidied up the graphs of battery level, heart rate etc.

Thank you for the testers for helping to identify these issues so we can fix them before we make this version generally available.

Note that the reason for producing this new version is to allow us to use the new Machine Learning (“AI”) seizure detection algorithm that has been devloped from user contributed seizure data. The intention of the Machine Learning (“AI”) algorithm is that it should give better seizure detection performance with fewer false alarms than the current OpenSeizureDetector algorithm. If you would like to help test the app before we make it available to all users, please email graham@openseizuredetector.org.uk with your google email address – then I can add you so that Google Play Store will give you the test version of the app.

Thanks

Graham.

A new draft of the Version 5.0 release (V5.0.0f) is now available for testers (you need to ask for access to this at the moment, because it is not quite ready for ‘production’ use.

The changes in this draft are:

• Fixed an issue with the whole system crashing on older android devices when the machine learning model is activated – it should now warn you that the model is not compatible with your hardware rather than crashing.
• Improved the Network Data Source (which is used to give remote alarm notifications using your local wifi network) so that you can see the results of the machine learning algorithms on the remote devices.

The intention of the Machine Learning (“AI”) algorithm is that it should give better seizure detection performance with fewer false alarms than the current OpenSeizureDetector algorithm. If you would like to help test the app before we make it available to all users, please email graham@openseizuredetector.org.uk with your google email address – then I can add you so that Google Play Store will give you the test version of the app.

Thanks

Graham.

Thanks to everyone who has contributed to our Data Sharing programme, we now have enough seizure data to start developing a Machine Learning (‘AI’) model to replace the deterministic seizure detection algorithm that OpenSeizureDetector uses.

We now have a model that has slightly better seizure detection reliability and improved false alarm performance, compared to the original algorithm, so is good enough to deploy for real seizure detection use (I’ll do a separate technical post about this later).

To implement this model, we need an update to the Android App. I am calling the updated version Version 5, because there are some very significant changes that I have made to it at the same time. The changes include:

• User Interface:
  • A new setup ‘wizard’ to guide new users through setting up the system – particularly for PineTime watches.
  • Simplified the settings screens.
  • Simplified the user interface and updated it to use modern ‘Material Design’ standards.
  • Seizure Detection:
  • The ability to download multiple Machine Learning (ML) models and combine the results for seizure detection, and prompt the user if a new recommended model is available for download.
  • Option to have the enabled seizure detection algorithms vote to decide whether to raise an alarm or not, to improve performance compared to a single algorithm.
  • Behind the Scenes
  • Improved reliability of data sharing reconnection and app startup/shutdown.
  • Split the code into various directories to make it clearer what does what.
  • Implemented a lot of the code modernisation proposals from user @AroonPro.

Testers Needed

There have been a lot of changes in this version of the app, so it needs a period of testing before we publish it as the main version on Google Play Store. There are three aspects to the testing:

• User Interface / User Experience: Does the user interface make sense? Does the initial setup ‘wizard’ work for different data sources and algorithms? What would make it more intuitive for new users?
• Functionality: Check that all of the functionality that you use in the current version still works the same (e.g. Mute function, SMS alerts, data sharing)
• Seizure Detection Reliability: I will re-run the automated tests using the seizures in the OpenSeizureDatabase (Data sharing database) to check that the app is performing as expected (the original OSD algorithm works as the previous version, and the new Machine Learning algorithm performs as well as it did on the computer version).
  
  If you would like to help with testing, please email graham@openseizuredetector.org.uk and we will work out a plan of how to do it.

A Few Screenshots below: