Diagnostic Imaging Europe (FEB/MARCH, 2020) on Care Mentor AI

Diagnostic Imaging Europe (FEB/MARCH, 2020) on Care Mentor AI

Why do Care Mentor AI artificial intelligence technologies outperform other pathology detection algorithms as well as radiologists?

Read the article in the FEB/MARCH 2020 issue of journal Diagnostic Imaging Europe by the Head of Research and Development of Care Mentor AI, Ph.D. Dmitry Blinov on the promising approach Care Mentor AI. The article was prepared jointly with colleagues from I.M. First Moscow State Medical University Sechenov. The article is devoted to the description of the advantages and prospects of computer vision technologies in identifying pathological changes and foreign bodies on chest radiographs.

The developers of Care Mentor AI proposed an original computer vision system, which is a binary classification model that can determine whether a patient has lung pathology or not. Two more models were used to detect and visualize foreign bodies on X-ray images of the chests. Test results showed that both the foreign body detection model and the classification model yielded reasonably good results.

Chest x-ray is the most common radiation diagnostic method in the world. It accounts for up to 45% of all x-ray studies. The wide availability of chest radiography is due to its low cost and high diagnostic potential in relation to such socially significant pathologies as tuberculosis, lung cancer, pneumonia.

However, chest x-ray is an example of diagnostic uncertainty, because the image is formed as a result of the overlap of anatomical areas with different structure, composition and density. The image can contain dozens of signs of pathological processes and conditions, while the patient is healthy. Of course, this makes it difficult to read and correctly interpret chest x-rays. Thus, differences of opinion between radiologists often lead to unreasonable additional examinations of patients.

Neural networks have the advantage of accurate interpretation and speed not only over other algorithms, but also over radiologists. The subsequent use of computer vision to identify pathological conditions in x-ray studies has allowed the development of high-performance models.

Read the full article here (p. 20):