| Summary: | The lack of universal (generalized) data sets, as well as the lack of annotated data, creates the need to study the possibilities of neural network approaches for specific data sets. The importance of building algorithms for detecting extrapulmonary pathologies on chest X-ray images is dictated by the great social significance of many diseases of this group (for example, cardiovascular diseases), given the availability of such images, due to the widespread use of minimally invasive and relatively cheap X-ray diagnostic methods. One of the most impor tant issues in solving the problems of automating the classification of medical images is data preparation. As a result of work on the image base, the performance of the final algorithm has been increased from 75 to 95 %. The processing of the entire volume of the obtained images and their diagnostics for a wide list of pathologies are difficult for medical institutions because of the limited resources. In this regard, it is advisable to use the automation of segmentation and recognition processes, which even at the first stages of development of the technology makes it possible to redistribute the attention of doctors, focusing on potentially pathological cases and returning attention to cases mistakenly identified as non-pathological.
|
|---|