Artificial intelligence detects rare genetic diseases on the face
The Cornelia de Lange syndrome is very rare, only about one out of approximately 45,000 people is affected. The disease often goes hand in hand with an intellectual disability. Also externally, the syndrome makes itself felt: The Affected usually have dense, curved eyebrows that have grown over the root of the nose, a wide, flat nose with forward-facing nostrils, a thin upper lip, down-drawn mouth and a small chin.
Doctors diagnose the disease so far, mainly on the basis of the outer appearance of the patient. Because the disease is so rare, not always easy. Now researchers have developed Software that can detect, on the basis of portrait photos, especially of children reliable rare hereditary diseases.
The Software recognizes 200 genetic diseases
Have developed the program called deep form of scientists from the USA, Israel and Germany, as they report in the journal “Nature Medicine”. It is more than 200 mostly very rare syndromes recognize the prepare, often in childhood problems. In the case of an initial suspicion, the Software limit the number of possible genetic causes and accelerate the diagnosis, says Peter Krawitz from the University hospital Bonn.
If you combine all rare hereditary diseases, occur frequently: Approximately two to eight percent of the population had a genetic syndrome, says Krawitz. A third to half of these diseases go hand in hand with mental impairments, which are often already in infancy.
“Because of the large number of possible syndromes, and its rarity, the correct diagnosis is a lengthy and expensive process,” writes the Team of Yaron Gurovich from the Boston-based company FDNA. So far, only a few experts unusual appearance could recognize pictures, or extremely rare the symptoms.
For the victims and their families, which often means a years – to decades-long doctor’s Odyssey, the constant search for the cause of the problem, new therapy trials, shattered hopes. Intelligent programs could improve the Situation significantly, so the idea of the researchers.
The Software deep the shape of investigated frontal shots of faces on characteristic abnormalities, and analyzes the shape of the eyes, the mouth, the chin or the distance between the eyebrows. Overall, the face detects 130 points on the face and compares these with 216 syndromes.
It works similar to a neural network that recognizes certain patterns in the face. “In the database, the patient is matched to a photo with lots of pictures, and a total similarity is calculated,” says bioinformatician Krawitz.
Training with more than 17,000 images
The most elaborate was, according to Krawitz, the Training of the Software to a data set of more than 17,000 images. Thus, the program practiced the recognition of Cornelia de Lange syndrome on the basis of 614 images of Affected and nearly 1100 pictures of other people. In a subsequent Test of whether a person has this syndrome or not, has reached the program has a reliability of 97 percent.
In Angelman syndrome, for which Training images of nearly 770 people Affected and about 2700 other people were used, the reliability of 92 percent. The Affected often have a small head that is flattened at the back, and a rather large mouth with protruding upper jaw.
In these two Tests, it was initially only a question of whether someone has this syndrome or not. In two further Tests, the researchers tested then, as well-deep shape to be able to assign a face photo, one of 216 different genetic defects.
The program provides only a presumptive diagnosis
According to the analysis, the System gave a Top Ten list of possible diagnoses: The probability that the actual genetic defect was to be found in this Top Ten, was about 90 percent. About 65 percent of the cases, the most likely current diagnosis even met.
Used the Software could be of children’s doctors, would those parents with eye-catching children, says Krawitz. However, the program was trying to deliver a suspected diagnosis, which would then need to be laboratory verified. In addition, only those genetic defects can be detected, can also trigger significant morphological variations of the face shape – not all hereditary diseases bring.
“The value lies in the fact, that some children, doctors can make in the ideal case, in consultation with a human geneticist, targeted diagnostics, Vera,” said the bioinformatician. So even the experts are overwhelmed with the diagnosis, often because of the number of newly discovered syndromes increases continuously.
The detection of genetic diseases is only one of many applications of Artificial intelligence in medicine. Similar software to be developed according to Krawitz for the analysis of other images, such as MRI scans or photos of the retina.