The agent-oriented system for interpretation chest X-ray images

The interpretation of X-ray images requires a significant amount of time, a great deal of experience, and a highly qualified radiologist. The solution to this problem is the implementation of artificial intelligence methods. This paper presents an intelligent system for interpreting chest X-ray images, which is based on an agent approach, a modular architecture, and multimodal reasoning. Work purpose – development of an open-source agent-based platform for comprehensive interpretation of chest X-ray images, capable of solving a wide range of X-ray diagnostic tasks with explicit and accessible reasoning at every step. Results. The classes of diseases were selected, the concept and architecture of the system were developed, the logic of decision-making and interpretation of results was formulated, and web applications were developed for interaction with the system. When testing the system on the CheXBench benchmark, the accuracy of interpretation reached 68.1%, which exceeds the accuracy of known competitive solutions. Practical significance. The system is intended for use as an AI radiologist assistant, as well as for training medical personnel. The system does not simply generate a diagnosis, but, in analogy with medical interpretation, conducts a step-by-step analysis with explicit and accessible reasoning, which makes its work understandable and predictable.