Around 50 million people suffer from epilepsy--about 1 percent of the world's population, say Forrest Sheng Bao at Texas Tech University, in Lubbock, and a few pals.

But diagnosing the condition is tricky. The gold standard is a recording of the electrical activity during a fit as measured by video and electroencephalogram (EEG) data. Of course, this kind of data is tremendously hard to get because of the disruption it causes to patients' lives. Imagine wearing the necessary electrodes and being within sight of a video camera continuously over a period of days, or even weeks.

In addition to this, some 70 or 80 percent of sufferers live in the developing world, where these kinds of measurements are even more impractical.

Bao and colleagues have come up with a system that may have a dramatic impact. Various groups have attempted to automate the process of epilepsy diagnosis using pattern recognition programs to spot the characteristic signature of the condition in EEG data. But these all depend on the EEG-video data that is so hard to get.

Bao and Co. have come up with a way to automatically diagnose epilepsy using data from recordings taken between fits, called interictal data. Obviously, this data is much easier to take.

The team developed the system by training a neural network to recognize the characteristic patterns in interictal data that indicate that the patient is epileptic. And the researchers claim an accuracy rate of 94 percent--about the same as experienced human operators, who usually have to strip various kinds of noise and artifacts out of the data before they can do their job.

That looks like impressive work that could have a major impact on the way that the disease is handled, particularly in the developing world.

Ref: arxiv.org/abs/0904.3808: Automated Epilepsy Diagnosis Using Interictal Scalp EEG