Facing up to diagnosis

The computer has been trained to spot facial traits

Scientists have developed a program to train computers to scan faces and spot genetic syndromes.

If this works we would be delighted

A spokesman for the Cornelia de Lange Foundation

They hope that by looking for specific facial traits the computer will help aid doctors with little experience of the conditions to be able to diagnose them.

Using a series of two dimensional full face photographs the computer has been trained to spot rare conditions like Cornelia de Lange, Fragile X and Williams-Beuren syndromes.

Charities are now hopeful that the computer system will help doctors spot conditions at a much earlier stage.

Diagnosis

The computer program represents the face in terms of a graph with 48 points. By comparing the position of these and the distances between them with those on faces in the data base the program is able to recognise conditions.

At first, the system was only 60% accurate, but then it was retrained to pay more attention to the eyes, nose, mouth and chin and the diagnosis rate shot up to 76%.

Previous studies attempting to identify syndromes using facial features have proved less reliable because they have not accounted for the whole facial make-up.

Sufferers with Cornelia de Lange syndrome generally share common characteristics which include slow growth, small stature, excessive body hair, and upper limb and heart defects.

They also tend to have thin eyebrows that often meet in the middle, long eyelashes and a short upturned nose.

Those with Williams-Beuren syndrome, suffer from a range of symptoms including heart defects, mental deficiency and high levels of calcium in infancy.

They tend to have wide mouths, full lips, small chins and puffiness around the eyes.

Those with Fragile X syndrome tend to have learning and behavioural problems and as they get older they generally have largish heads and prominent ears.

Although these facial characteristics can become more pronounced with age they can be quite difficult to spot in babies.

Accuracy

A spokesman for the Cornelia de Lange Foundation said they welcomed the discovery and looked forward to seeing how it could be adapted.

"We would be very interested in that anything helps with accurate diagnosis as early as possible and if this works we would be delighted.

"It would be interesting to see if this works on diagnosing the milder forms of the condition."

Dr Dagmar Wieczorek, of the Institute for Human Genetics at the Universitaetsklinikum, in Essen, Germany, fed in 55 photos of people with the conditions and found the computer accurately predicted 76% of cases.

She told the European Society for Human Genetics, in Strasbourg: "It takes a particular skill to extract a disease specific pattern from a facial appearance that is also influenced by family background and environment.

"The idea is to help inexperienced clinicians to make more accurate diagnoses."