Transformer Networks for Morphological Analysis and Functional Status Classification of Blood Cells

This study presents a groundbreaking AI-driven framework for the automated classification of blood cells, with a novel focus on determining the activity status of white blood cells (WBCs) through advanced morphological analysis. WBCs are pivotal to the immune system, and their activity status is a critical biomarker for diagnosing immune disorders, infections, and inflammatory diseases. Traditional methods for analyzing WBC activity are labor-intensive, subjective, and often inaccessible in resource-limited settings. To address these challenges, we propose an automated transformative two-stage learning architecture that integrates the strengths of Transformer Neural Networks (TNNs) [6] and pre-trained Convolutional Neural Networks (CNNs), such as AlexNet and RetinaNet Keras [11]. In the first stage, spatial features are extracted from blood cell images using CNNs, capturing intricate structural details essential for accurate analysis. These features are then processed by a Transformer Network, where self-attention mechanisms enable sophisticated classification, identifying not only cellular anomalies but also the functional status of WBCs. This unique combination of CNNs and TNNs leverages the former's robust feature extraction capabilities and the latter's ability to model complex relationships within data, resulting in unparalleled diagnostic precision. The proposed system achieves outstanding performance metrics, including high accuracy of 0.99, recall of 0.99 and F1 scores of 0.985, demonstrating its reliability for real-world clinical applications. By automating a traditionally manual and errorprone process, this framework significantly enhances diagnostic efficiency and accessibility, particularly in underserved regions. This work represents a major advancement in hematological diagnostics, offering a scalable, rapid, and accurate solution for improving patient care in diverse healthcare settings.