Dogs can predict epileptic seizures in their owners: sensors can convert that prediction into action

Assistance dogs are often used by those with physical conditions to support their daily lives, and we know that they become highly attuned to their owner’s behaviours. Some dogs are known to be able to detect the onset of an epileptic seizure by their owner and can be trained to communicate this detection to the outside world.  They predict seizure onset through the detection of Volatile Organic Compounds (VOCs); chemicals released by the skin that the dogs can smell. If the dog has been trained to respond in a particular way to relevant VOCs; ie by spinning around or rolling over; that movement can be ‘read’ by a collar sensor and reported directly to the owner or carer via another device.

Professor Alan Smeaton and the team at Insight DCU are using wearable sensors on service dogs to convert predictive behaviour into real time reporting.  When we apply machine learning and other AI techniques to raw data from wearables then we can get higher level interpretations such as step counts, caloric expenditure or recognition of activity types like walking, running, swimming, etc.  In recent work Prof Smeaton has used wearable accelerometers on the collars of trained dogs and used the data to automatically classify the dog’s activities into spinning or rollovers, exactly the kind of signalling behaviour used by assistance dogs. The accuracy of the automatic detection is already quite good but could be improved further.

The next stage is to combine simple sensors worn by assistance dogs to automatically detect their trained signalling behaviours to their human charges, in particular to the onset of an epileptic seizure and to do this in real time.