INTELIGENCIA ARTIFICIAL EXPLICABLE EN EL LABORATORIO CLÍNICO: OPORTUNIDADES Y BARRERAS PARA MEJORAR LA PRECISIÓN DIAGNÓSTICA Y LA SEGURIDAD DEL PACIENTE

Abstract

ABSTRACT Artificial intelligence (AI) has been progressively incorporated into clinical laboratories as a tool to support data processing, result interpretation, and diagnostic decision-making. However, the use of complex models has raised concerns regarding transparency, interpretability, and the ability to supervise algorithmic decisions, particularly when these decisions may influence patient care and safety. In this context, explainable artificial intelligence (XAI) has gained relevance by providing mechanisms intended to facilitate the understanding of predictions generated by AI models. This review aimed to analyze the opportunities and barriers associated with the application of XAI in the clinical laboratory, considering its potential contribution to diagnostic accuracy and patient safety. A descriptive and analytical bibliographic review was conducted of scientific literature published mainly between January 2021 and September 2026 through searches in PubMed/MEDLINE, Scopus, Web of Science, ScienceDirect, SpringerLink, and Google Scholar. Studies addressing explainable artificial intelligence, laboratory medicine, diagnosis, accuracy, validation, patient safety, and interpretability were considered. The reviewed evidence showed potential applications of XAI across the preanalytical, analytical, and postanalytical phases, as well as in hematology, biochemistry, microbiology, immunology, and digital pathology. The main opportunities included increased model transparency, traceability of predictions, support for error identification, and strengthened professional oversight. The main barriers involved data quality and heterogeneity, interoperability, external validation, bias, information protection, the lack of standardized explainability criteria, and the need for professional training. It was concluded that XAI represented a complementary tool with potential to support diagnostic accuracy and patient safety, provided that its implementation was accompanied by rigorous validation, continuous monitoring, human oversight, and appropriate ethical and regulatory criteria.