The neural network was taught to recognize 216 rare hereditary diseases from a photo.


Researchers have developed an artificial intelligence system that allows you to accurately diagnose 216 rare hereditary diseases from a photograph. As reported in Nature Medicine , she was taught to recognize a genetic disorder (choose from the 10 most likely options) with 91 percent accuracy. Scientists also simplified the application of the system in practice: they created a mobile application for medical professionals, which allows you to determine the genetic disorder from a photograph of a patient.

Y.Gurovich et al. 
/ Nature Medicine, 2019

It is often difficult to diagnose a hereditary disease. There are several thousand diseases associated with genetic disorders, most of which are extremely rare. Many doctors during their practice may simply not be confronted with similar diseases, therefore a reference computer system that would help recognize rare hereditary diseases would facilitate diagnosis. Researchers have already created similar systems based on facial recognition, but they were able to determine so far, no more than 15 genetic disorders, while the accuracy of recognition of several diseases did not exceed 76 percent. In addition, such systems sometimes could not distinguish a sick person from a healthy one. At the same time, the training sample often did not exceed 200 photos, which is too small for deep learning.

Therefore, American, German and Israeli scientists and employees of FDNA, under the leadership of Yaron Gurovich from Tel Aviv University, developed the DeepGestalt facial recognition system, which made it possible to diagnose several hundred diseases. Using convolutional neural networks, the system divides the face into separate fragments with dimensions of 100 × 100 pixels and predicts the probability of each disease for a particular fragment. Then all the information is summarized and the system determines the likely disorder for the person as a whole.

DeepGestalt splits a face in a photo into separate fragments and assesses how closely they correspond to each of the diseases in the model. 
According to the aggregate of fragments, the system makes a ranked list of possible diseases.
Y.Gurovich et al. 
/ Nature Medicine, 2019

Researchers have trained the system to distinguish a specific hereditary disease from a number of others. For training, they used 614 photographs of people suffering from Cornelia de Lange syndrome – a rare hereditary disease that manifests itself, including in the form of mental retardation and congenital malformations of internal organs. As a negative control, the authors used more than a thousand other images. DeepGestalt distinguished Cornelia de Lange syndrome from other diseases with an accuracy of 97 percent (p = 0.01). The authors of other studies failed to achieve 87 percent accuracy, while experts put the correct diagnosis, on average, 75 percent of cases. In another experiment, scientists used 766 photographs of patients.Angelman syndrome (“Parsley syndrome”), which, inter alia, is characterized by chaotic movements, frequent laughter or smiles. The system recognized the disease with an accuracy of 92 percent (p = 0.05); in a previous study, the determination accuracy was 71 percent.

The researchers also taught the system to recognize different types of the same hereditary disease using the example of Noonan syndrome . There are several types of this disorder, each of which is caused by mutations in a particular gene and each has slight differences in facial features (for example, rare eyebrows). Using a sample of 81 photographs, the authors of the article taught the DeepGestalt system to distinguish five types of this disease with an accuracy of 64 percent (p <1 × 10 -5 ).

In total, scientists used a total of 17,106 photographs, representing 216 inherited diseases, for training the system. The effectiveness of DeepGestalt work was checked by researchers on 502 photographs of patients who had already been diagnosed, and on another sample of 329 photographs of patients with a known diagnosis from the London medical database. The system determined the patient’s disease from the 10 most probable variants with an accuracy of 91 percent (p <1 × 10 – 6 ).

The researchers also made it easier for DeepGestalt to be used in practice – they created a phenotype diagnostic platform for hereditary diseases, as well as a Face2Gene mobile app for doctors  , with which the doctor can diagnose his patient.

Last year, researchers created a system for automatic recognition of plants from their images in herbariums. The convolutional neural network learned to identify plants with 90 percent accuracy.

Related Articles

Leave a Reply

Your email address will not be published. Required fields are marked *

Back to top button