What to do if a person lives in the “outback” and needs the advice of world-class doctors? How to diagnose if the disease is rare? Scientists at the NTUU “Kyiv Polytechnic Institute named after Igor Sikorsky” believe that artificial intelligence (AI) can help diagnose patients with lung disease. They are developing an AI platform that will analyze X-rays, CT scans or MRIs of the lungs in seconds and help establish the correct diagnosis.
Mr. Serhiy has been working in the field of artificial intelligence for many years, and six years ago, thanks to an accident, he became interested in the possibility of diagnosing lung disease with the help of this technology. “A colleague from the scientific community were misdiagnosed by doctors”, he explains. – We decided to test him using artificial intelligence algorithms. It was found to be a rare immune disease that the doctors who performed the examination did not encounter at all. And so arose the idea, figuratively speaking, to “put” artificial intelligence in a medical gown”.
The problem of incorrect or too late diagnoses is not only related to the health care of Ukraine. According to the scientist, there is a lack of medical professionals who could recognize lung disease at an early stage in many countries.
– Scientists in many countries of the world are working today to create AI systems for establishing medical diagnoses, – Serhiy Hryhorovych continued. – And there are already good results.
Prototype of the human brain
What lung diseases will the platform created by Kyiv scientists be able to recognize? And how fast?
– To establish the diagnosis, today a person needs an appointment with a therapist, who gives a referral for an X-ray or CT scan, then the radiologist processes the images, – said Mr. Serhiy. – And only after two or three days the patient gets an appointment, coronavirus inflammation, for example, can develop instantly and take the patient’s life in a few hours, and our platform identifies signs of 14 lung diseases (oncology, tuberculosis, coronavirus inflammation, pneumonia and less common diseases) in a few seconds. Of course, treatment should be prescribed by a doctor, but the AI system will save patient’s precious time.
The neural network is a prototype of the human brain. The AI model learns from images from open medical databases, reads descriptions, forms and remembers the signs of various lung diseases. More than 300,000 images have already been analyzed, and no doctor can have so many clinical cases in their entire practice.
In order to make sure that the model works correctly, scientists tested it. The results obtained with the help of the model were compared with the results of the analysis of the world’s leading pulmonologists – the AI system turned out to be several percent more accurate.
Tips from practitioners
The project is carried out by specialists of the Department of Computer Science of Kyiv Polytechnic. “I would like to note that this is one of the best artificial intelligence teams in Ukraine, said the scientist. – It includes both experienced professors and young scientists”.
– Practitioners are involved in the formation of the national DataSet (reference data set), testing the interface of the platform, – said Mr. Serhiy. – It should be as convenient and clear as possible! Because IT professionals sometimes create systems that no one but themselves understands (laughs, ed.). We want the development to be used by doctors without a need to look for an IT specialist who will help to understand «what and why».
The system will work “in the cloud” and to make it work as efficiently as possible, a cloud service provider was invited to the team. “Doctors from the most remote settlements of the country will be able to enter the system and compare their diagnosis with the findings of artificial intelligence”, said Serhiy Stirenko. “It’s like inviting the best pulmonologists to a consultation in a remote village”.
The project should be completed in December 2021. To date, the platform has been created and practitioners give the latest recommendations for its convenience. The creation of a Ukrainian DataSet in subcontractor clinics is also underway. “We plan to teach the system on lung images of patients from Ukraine”, the scientist added. “This will be Ukraine’s contribution to world science, which deals with lung diseases using artificial intelligence algorithms”.
Upon completion of the project, the scientists plan to provide access to the platform to all interested hospitals from all over Ukraine.
Svitlana GALATA
