Justification for selecting the neural network type for inclusion in the architecture of the developed medical information system

Abstract

This article provides a brief overview of the application of neural networks in medical systems for disease diagnosis. The relevance of developing a medical information system (MIS) with artificial intelligence for otolaryngologists is justified. The developed automated workstation (AWS) for doctors is presented. The MIS AWS enables the following tasks: loading, storing, and viewing examination results in the DICOM format. The MIS is a web platform with a client-server architecture, utilizing technologies such as PostgreSQL, Python, Django REST Framework, Docker, Docker-compose, and Vue.js. Vue.js was chosen as the framework for Frontend, along with the libraries Vuex and Vue-router. The MIS consists of four modules: Vue.js framework, Django framework, database (DB), and file system (FS). The information-communication structure for the exchange of medical images is presented, where the MIS serves as the core, facilitating the overall data preparation and accumulation cycle for CT scan analysis. Based on the analysis of neural network architectures for medical image recognition, the decision was made to implement a convolutional neural network (CNN) into the MIS structure. The automated centralized repository for medical data on patient examinations currently performs the function of accumulating and storing information for the training and testing of the neural network.