The finding, hailed by experts as a "major milestone", has the potential to dramatically improve the independence and treatment of millions of people living with the condition globally.

Melbourne researchers reported in the Lancet Neurology medical journal on Thursday that a small device can be implanted in the brain to monitor electrical activity which can change before seizures occur.

The technology, developed by US company NeuroVista, involves electrodes being surgically placed between the brain surface and the skull. The electrodes feed data to a device implanted under the skin of the user's chest which then transmits the information to a wireless hand-held device that calculates the probability of a seizure.

Lead author of the report and a neurologist from the University of Melbourne, Professor Mark Cook, said the hand-held device which can be carried in a handbag or pocket has three coloured lights to warn patients of a low, moderate or high risk of an impending seizure. It also beeps and vibrates to alert people to the results.

He said the device, which was trialled in 15 people with focal epilepsy, predicted seizures with the highest level of warning 56 per cent to 100 per cent of the time for eight of the patients who used it for more than four months. The time between the alert and the seizure varied from eight minutes to four hours, with the average time being about 100 minutes.

Professor Cook said this was life-changing for some patients who had previously lived with the uncertainty of not knowing when seizures would strike. In some cases, the device meant they can warn people around them.

"Knowing when a seizure might happen could dramatically improve the quality of life and independence of people with epilepsy and potentially allow them to avoid dangerous situations such as driving or swimming, or to take drugs to stop seizures before they start rather than continuously as at present," he said.

While more than 60 million people have epilepsy, about one-third are unable to control seizures with existing treatments.

The small proof-of-concept study involved 15 people aged 20 to 62 years who were having between two and 12 seizures per month and had not had their seizures controlled by at least two anti-epileptic drugs.

After having the device implanted, one month of data was collected on participants' seizures so it could be analysed to construct an individual algorithm of seizure likelihood.

During the first four months after implantation, the device correctly predicted seizures at least 65 per cent of the time.