Classification of professional computer software, including artificial intelligence-based applications, used in healthcare and clinical medicine

Vollständiger Abstract

Worum geht es in dieser Arbeit?

Background. Software applications based on artificial intelligence (AI) technologies and computer vision programs (CVPs) are increasingly being used alongside conventional software tools across various areas of medicine, from prevention and diagnosis to rehabilitation and healthcare system management. The rapid growth in the number of such programs and the absence of a unified classification system impede their comparison, informed selection for implementation, regulatory assessment, and scientific analysis. Objective: To develop and substantiate a classification framework for software applications, including those based on AI and CVPs, used in clinical medicine and healthcare, covering their application context, implementation features, data characteristics, software architecture, and training parameters. Material and methods. The experience in the development and clinical implementation of software applications, including those based on AI CVPs for visual identification and differential diagnosis of skin neoplasms, was analyzed. On this basis, classification criteria were identified and grouped into five thematic blocks. Results. A system based on the following 19 classification criteria is proposed: by user, organizational level of application, purpose of use, field of medical application, and type of medical care (application context block); by deployment location, integration into equipment, and program type (deployment and implementation block); by data type, input data modality, and equipment used (data and equipment block); by number of models, number of classes, and image processing method (architecture and processing block); by type of training, number of training objects, technical and professional training of data (training characteristics block). For a number of features, ordinal gradations of the conditional level of complexity and reliability are proposed. Conclusion . The proposed classification provides a unified framework for describing software, including those based on AI and CVPs in clinical medicine and healthcare, facilitates their comparison and selection, and can also serve as a basis for methodological and regulatory requirements for such programs.