O Uso da inteligência artificial como potencial no diagnóstico de leucemia mieloide aguda

Abstract

Introduction: Acute Myeloid Leukemia (AML) is a cancer with high clinical, morphological, and molecular heterogeneity, characterized by the ≥20% of myeloblasts in peripheral blood or bone marrow. Thus, cytogenetics, immunophenotyping, and molecular genetics are widely used in diagnosis. However, due to the limitations of these approaches, studies have applied artificial intelligence (AI) to analyze laboratory images, enabling faster and more accurate diagnoses and medical decision-making. Objective: To investigate the potential of artificial intelligence applied to diagnose acute myeloid leukemia. Methodology: This is a systematic literature review. We searched the PubMed/Medline and EMBASE databases using the keywords diagnosis, machine learning, artificial intelligence, and acute myeloid leukemia, covering publications between 2010 and 2024. Results and Discussion: This study highlights the effectiveness of AI techniques in diagnosing AML, as well as Convolutional Neural Networks, which obtained an area under the curve AUC of up to 97% in image analysis, while Linear Discriminant Analysis achieved 94% accuracy in identifying myeloblasts. Multiparameter Flow Cytometry showed high consistency with manual analyses (kappa = 96.3). Simplified models, such as Artificial Intelligence Prediction of Acute Leukemia, achieved an AUC of 89% in low-infrastructure environments, advanced techniques, such as Random Forest combined with Support Vector Machine and Convolutional Neural Networks, achieved up to 99.8%, with emphasis on pediatric diagnostics. On the other hand, the Least Absolute Shrinkage and Selection Operator and Gradient Boosting Trees regression showed high accuracy and efficiency in risk stratification and survival prediction, with AUC of 99% and 80-85%, respectively. Conclusion: AI offers benefits such as increased accuracy, standardization, and diagnosis agility but faces challenges related to data quality, clinical integration, and professional acceptance, underscoring the need for external validation and continuous training.