Systematic review on computer vision techniques applied to leukocyte classification

Abstract

Introduction: Automated leukocyte classification using computer vision is a promising alternative to manual analysis, which is time-consuming and subjective. Objective: This systematic review aims to analyze and synthesize evidence on machine learning algorithms applied to leukocyte classification in peripheral blood smear images, evaluating their performance and challenges. Methods: A systematic review (PRISMA) was conducted in MEDLINE/PubMed, Embase, and Scopus for studies published between 2020 and 2025. Results: Out of 300 initial records, 28 studies were included. The findings indicate a predominance of deep learning models (CNNs, YOLO) with accuracies often exceeding 95% in the classification of mature leukocytes. Conclusion: Although technically mature for normal cell classification, the field faces challenges such as methodological heterogeneity across studies and a gap in the classification of immature and atypical cells, which have greater clinical relevance.