Network neurons diagnosed 216 rare genetic diseases

US, German and Israeli scientists have collaborated to develop the DeepGestalt neuron network and Face2Gene mobile application, allowing doctors to identify genetic disorders from a patient's image to 91% accuracy.

According to Nature Medicine, an international research group has successfully developed an artificial intelligence system that can accurately diagnose 216 rare genetic diseases through photographs.

The neuron network is taught how to recognize a genetic disorder (selected from one of the 10 most likely options) with an accuracy of 91%. Scientists have also simplified the application of the system in fact: they have created the Face2Gene mobile application , which allows doctors to identify genetic disorders from a photograph of a patient.

called DeepGestalt, helps diagnose hundreds of diseases through images - (Photo: Nature Medicine).

It is often difficult to diagnose genetic diseases. Literature knows a few thousand related diseases, most of which are extremely rare. Many doctors during practice may have never encountered such diseases, so a reference computer system that helps identify rare genetic diseases will facilitate diagnosis.

Researchers have created similar systems based on face recognition , but so far only identified no more than 15 genetic disorders and the accuracy of recognizing some diseases does not exceed 76%. . In addition, such systems are sometimes unable to distinguish patients from healthy people. At the same time, the training sample usually does not exceed 200 images, too little to learn deeply.

Therefore, American, German and Israeli scientists with the work of FDNA employees, under the guidance of Yaron Gurovich from Tel Aviv University developed the face recognition system called DeepGestalt, Help diagnose hundreds of diseases. Using convoluted noron networks, the system divides the face into separate pieces of 100 x 100 pixels and predicts the probability of each disease for a particular piece. After that, all the information is summarized and the system of identifying disorders is possible for an entire person.

In total, the scientists used 17,106 photos, representing 216 genetic diseases, to "train" the system. The efficacy of DeepGestalt was investigated by 502 researchers on 502 photos of patients diagnosed with the disease and on another sample of 329 diagnosed patient images from the London medical database. In 10 possible disease options, the system identified the disease with an accuracy of 91%.