ISSAI presents its End-to-End Deep Diagnosis of X-ray images service

Scientists from NU’s Institute of Smart Systems and Artificial Intelligence presented their project on End-to-End Deep Diagnosis of X-ray Images. This service is available on the Institute’s website.

The system is based on deep learning technologies, and analysis of uploaded images of pathologies is done completely online in a matter of seconds. This method allows medical workers to identify all kinds of abnormalities and potentially dangerous malformations. Scientists are confident that such a service will help doctors in the effective diagnosis of pathologies and, at the same time, reduce the cost of medical care.

“We have collected a set of medical imaging data, including images of chests, fingers, wrists, elbows, forearms, hands, humeri, feet, kneecaps, ankles and hips, as well as images of spine and teeth. After that, we have used the Nvidia DGX-2 supercomputer to store and process these images. The processing power of the DGX trained the model. The use of supercomputer trained the system to accurately determine the type of X-ray and pathology based on the details of the image, ” said Yernur Nurambek, ISSAI researcher and co-author of the project.

The service is available as a demo version on the Institute’s website: . To use the End-to-End Deep Diagnosis of X-ray Images system, the doctor needs to upload a scanned X-ray image on the website. The diagnostics process consists of several stages: image identification (“X-ray or Not X-ray”), identification of the image type (chest, shoulder, elbow, lower leg, etc.), and identification based on an individual deep learning model for each type of X-ray image. 

At the moment, the system can only detect the pathology of the chest, but in the future, the scientists plan to improve the system and add new deep learning models  to identify pathologies of other parts of the body.