Histological analysis, crucial for identifying pathological tissue changes, traditionally relies on immunohistochemical studies. In this process, thin histological sections are treated with antibodies to detect specific substances within tissues, such as proteins. Despite its widespread use, this method has limitations, including time consumption and potential for error, even by experienced histopathologists.

Developed by RUDN University doctors, the EndoNet neural network addresses these challenges by automating and enhancing the analysis of immunohistochemical staining of the endometrium. The model utilizes two key components: a detection module that identifies important segments of cell nuclei and a calculation module that determines the H-score, a numerical value reflecting the abundance of specific proteins in tissues.

Testing EndoNet on a dataset of 1780 endometrial histological samples yielded promising results, with an average accuracy of 77%. Particularly impressive was its accuracy in analyzing stromal tissues, surpassing 85%. This automated system not only accelerates the diagnostic process but also minimizes errors resulting from variations in immunostaining techniques and pathologists’ biases.

EndoNet offers a universal solution for histology departments, improving efficiency and standardization in pathology diagnostics. Its potential to aid in diagnosing conditions such as miscarriages, infertility, and unsuccessful IVF attempts underscores its significance in clinical practice. With its successful development and validation, EndoNet represents a significant step forward in leveraging artificial intelligence for precision medicine in pathology.

Article written by Scientific Project Lomonosov