[ad_1]
07:00 p
Wednesday, January 16, 2019
D
Researchers from more than one country said they have managed to develop a computer program that can identify rare genetic diseases in humans through its optical image.
American, Israeli and German researchers presented the program in the latest issue of Nature Medsine, called Deep Shape. According to them, the program could identify more than 200 diseases, most of which are rare syndromes.
According to its developers, the suspect can identify the possible genetic causes of the disease, which allows a quick diagnosis, said Peter Kravets, a program development participant, from the University Hospital Bonn in Germany.
At the same time, however, the researchers noted that their program had many other capabilities.
Between 2 and 8% of humans suffer from a rare genetic syndrome, he said, adding that one-third to one-half of these illnesses coincide with mental retardation, which already appears in the US. ;childhood. "Because of the large number of rare syndromes, for a long and very expensive diagnosis," according to the team of researchers under the supervision of Yaron Gorovic, of the American company FDNA.
Until now, only a few experts have been able to identify unusual forms or extremely rare symptoms of the disease, but automated systems have the potential to improve this situation, researchers said. .
The Deep Shape program examines face facial images for any striking and distinctive symptoms of certain diseases: badysis of the shape of the eyes, mouth and chin as well as the distance between the eyebrows.
The artificial neural network, a technique similar to that of the brain, reproduces motifs, including 130 facial points, corresponding to syndrome 216. Depending on the extent of compatibility or disagreement with these points, technology provides a list of genetic causes the most likely of any of these rare diseases ", then compared in the database between these many images and reaches a similarity with the total", according to Kravets, specialist in bioinformatics.
According to Kravets, the most stressful phase for researchers during the development of the program was to train the program on a collection of over 17,000 images, particularly to identify the Cornella syndrome of lang. And nearly 1100 other photos.
At a later test to determine if a person had this syndrome, the program achieved a confidence index of 97%.
The degree of confidence in the program was 92% in the identification of the Angelman syndrome, formed by the program through about 770 wounded and about 2,700 others.
However, the researchers did not investigate whether a person was infected with the syndrome, but both tests were designed to control the sensitivity of the software.
In two experiments, the researchers tested the program's ability to detect one of the 216 genetic disorders of one of the presented images.
After the examination, the new computer system presented ten possible key diagnoses of the images that he was reading.
The probability of having a genetic disorder among these 10 diagnoses is already 90%.
The diagnosis, considered the most likely program in about 65% of the cases presented, was correct.
The researchers hope the program will be used in children's clinics attended by parents who complain that their children are suffering from obvious symptoms, Kravets said.
However, the program provides only possible diagnoses, leaving the labs the last word, to confirm the extent of genetic disorders.
According to Kravets, the interest of the program is to help pediatricians coordinate with a human genome specialist to make a diagnosis going in a particular direction, particularly in the face of the considerable efforts made by experts involved in the diagnosis of rare cases because of the continuous increase of syndromes of rare diseases detected. Day by day.
It should be noted that the diagnosis of genetic diseases is only one of the many applications of artificial intelligence in medicine.
According to Kravets, similar software is being developed for use in the evaluation of other images such as magnetic resonance imaging or retinal images.
"The development of these programs is still in its infancy," Kravets said. These networks will be connected in a few years. "
Source link
