We can improve diagnostics with the help of AI - head of Filatov Institute department
The Head of the Department for the Study of Biological Effects and Application of Lasers in Ophthalmology at the Filatov Institute of Eye Diseases and Tissue Therapy, Andriy Korol, told the Interfax-Ukraine agency about the prospects of using artificial intelligence technologies in medicine.
Author: Anna Levchenko
- Why did the idea of using artificial intelligence technologies in the diagnosis of eye diseases arise?
- Nowadays, it is very difficult for any scientific team to make a significant breakthrough in any field of science. All significant achievements occur at the junction of several directions, disciplines and even sciences. For example, there are such serious and, unfortunately, very common diseases as diabetes, which affects at least 2 million people in Ukraine alone and at least 500 million people in the world. Diabetes is dangerous due to its complications, the spectrum of which is very wide. Ophthalmological diseases caused by diabetes are one of the most serious complications of diabetes, they can lead to complete loss of vision. Ophthalmologists know such patients well, and endocrinologists, in turn, understand that the involvement of ophthalmologists is mandatory.
At the same time, we understand that endocrinologists cannot examine the fundus. Ophthalmologists can examine the fundus and describe its condition, but in the modern world, technologies are needed that can record "everything that everyone can," compare, analyze and help the doctor make a more accurate diagnosis.
- Are there already any implemented projects in this direction?
- Ophthalmology specialists of the Filatov Institute are already participating in the development of the Ukrainian startup CheckEye. It provides the technical side of joint work between endocrinologists and ophthalmologists.
- Let's return to the issues of application of artificial intelligence technology. What exactly does AI do - describes results, analyzes, diagnoses?
- It can do many things, but first it must be taught to see the signs of the disease of the fundus on digital photographs, in this case - a case of diabetic retinal damage. We teach artificial intelligence this on several thousand pictures, we explain to it that, for example, this looks like a hemorrhage, this looks like a vascular injury, and everything is normal in this photo. Then we offer it pictures of real patients suffering from diabetes, and the system, analyzing the detected signs, gives its conclusions. Of course, we control AI, correct it, it is fixed, that is, we train it in parallel.
- Am I correct in my understanding that after all AI establishes a diagnosis under the control of doctors?
- Definitely. This is the second level of AI training, which takes place under the control of a person, our experts, and ophthalmologists of the institute. We can improve diagnostics with the help of AI, and we can already see that in terms of effectiveness, collaboration between AI and humans surpasses, for example, an ordinary ophthalmologist.
- Has the project with endocrinologists (diagnosis of diabetic damage to the eye fundus) already been launched? Is it working or still only a development and an idea?
- This project is already working in several regions, for example, in Chernivtsi, Zakarpattia, Kyiv. Of course, this is only the beginning, but it is already a working option that helps both doctors and patients.
- How does it work technically?
- Technically, it works like this. The idea of this system is to reduce the burden on specialists, that is, this system should work at a pre-medical level. In order to take a picture of the fundus, a doctor is not needed. A nurse, any person, who has been trained to work with special equipment, can do this. Moreover, it is quite inexpensive, it can be some kind of pharmacy, optician's store, and polyclinic. Even it can be a mobile team of doctors, which has a camera and a camera operator. The mobile team will be able to reach the most remote village in the mountains, where there is not even a family doctor. Of course, you need a laptop and Internet access.
The photo is uploaded to the cloud storage from anywhere in the country, processed and in 30 seconds a conclusion is issued as to whether a person has a diabetic lesion of the fundus or not. In this, the system is even trained to determine the severity of the lesion. Of course, after the examination, the patient should consult a doctor.
That is, one of the advantages of the system is that people who do not have fundus damage do not take the specialist's time for diagnosis.
- How many patients have already been examined using your system?
- Already more than 2,000.
- To what extent can you trust the conclusions and diagnoses made by artificial intelligence?
- AI does not establish a diagnosis - it determines the presence of a disease, that is, a deviation from the norm. In this regard, it is a very sensitive system, the accuracy is over 93%. These are very high indicators for diagnosis. To be honest, they even exceed the indicators of a human ophthalmologist.
- What other areas of application of AI technology do you see in ophthalmology?
- This is a very broad field. We have now worked out diabetic retinopathy, at the next stage we will develop in several directions. These can be specific eye diseases, retinal diseases, glaucoma, age-related degeneration, or other diseases - manifestations of diabetes. With the help of such a system, we can see the symptoms and signs of various diseases, for example, hypertension, vascular lesions, signs of hereditary diseases, and a number of rare diseases.
- How is the use of AI in medicine and ophthalmology developing in the world? Are Ukrainian developments now at the forefront or are we lagging behind?
- Based on what we observe at international congresses, I see that the Filatov Institute is in the vanguard. Indeed, in many countries, work is ongoing, developments are underway. There are many international projects. We cooperate with Poland, Romania, and Malaysia.
- When these systems can be launched as working?
- Well, look, it took us about two years to develop a system working with diabetes. Of course, those diseases that occur more often require less time, those that occur less often require more time, because a certain number of images must be collected for training the system. Since we have already developed some technologies, we can launch research on new nosologies every 1.5 years.