Artificial Intelligence in Hematology

Artificial intelligence (AI) has emerged as a powerful resource in healthcare for diagnosis and management. However, successful AI deployment depends on well-developed software for the preprocessing and analysis of digital medical images, along with deep learning algorithms for classification and interpretation. Deep machine learning is instrumental in identifying and interpreting anomalies within electronic health records or digital footprints. Over the last decade, AI has made significant impact in hematology, particularly in cellular, molecular, and genetic analysis. Notable examples include achieving an accuracy rate of over 95% for classifying peripheral smears and reporting a precision-recall rate exceeding 94% for detecting cellular-level features and segmentation tasks. When a multimodal diagnostic workflow was used, an accuracy of more than 95% was reported in differentiating subtypes of myelodysplastic syndromes. It's crucial to recognize that fundamental hematology diagnosis still relies on microscopic image analysis and the evaluation of flow cytometric two-dimensional plots, which may lead to interpretation inaccuracies. Nonetheless, significant progress has been achieved, including the development of foundation models, transformer-based architectures, and generative diffusion models, as well as the integration of omic data, which may help overcome existing limitations in the future.