PROTECTED INTELLECTUAL SYSTEM FOR ANALYSIS OF BIOMEDICAL IMAGES IN DIGITAL ONCOHEMATOLOGY

In the context of accelerated digitization of healthcare, issues related to the information security of medical data are gaining particular relevance. The article presents an approach to minimizing risks associated with the implementation of intelligent systems for processing biomedical images in oncohematology. Information security measures were integrated at the system design stage. It was developed in accordance with current regulatory requirements of the FSTEC and FSB of Russia in the field of protecting critical information infrastructure. It provides multi-level protection of patients' personal data and ensures the integrity of analysis results. To increase trust in the decisions generated by the artificial intelligence system, its interpretability was implemented: the doctor can visually assess which areas of the preparation the model focused on when classifying cells. This allows distinguishing a justified decision from a reaction to background artifacts and confidently verifying the system's recommendations. Risks of digital infrastructure failure were taken into account: an autonomous operation mode and a function for generating a paper report were introduced. Pilot implementation of the system allowed evaluating its impact on key stages of the diagnostic process. A particularly noticeable effect was achieved at the stage of annotating bone marrow preparations: the system significantly reduced the cognitive load on experts, simplified the identification of blast cells, and contributed to more consistent and complete marking of morphological elements. As a result, the sensitivity of the analysis increased–the number of missed blasts decreased, and doctors noted the high usefulness of AI recommendations. These results confirm that only a comprehensive approach, combining technical, organizational, and legal measures, allows creating a reliable and secure digital ecosystem that truly integrates into clinical practice.