Detection and morphological classification of neutrophils using convolutional neural networks to support left shift diagnosis

Abstract

A left shift is characterized by an increased number of immature neutrophils (band cells) in the bloodstream, indicating a bone marrow response to acute infection or inflammation. Leukocyte counting and the morphological differentiation between segmented and band neutrophils are crucial steps in laboratory screening. Although automated hematology analyzers are available, they are expensive and often face challenges in accurately distinguishing cell morphology, particularly in three-part differential (3-DIFF) systems. This limitation frequently generates system alerts (flags), requiring manual microscopic review to complement the analysis. However, manual counting is time-consuming and subject to interobserver variability. This study aimed to evaluate the feasibility of the YOLOv8 computer vision architecture for the detection and classification of neutrophils. A dataset of digital microscopic images obtained from the Kaggle platform was curated and annotated using Roboflow software, with the application of data augmentation techniques. The model was trained in a cloud-computing environment using Google Colab. The results demonstrated promising algorithm performance, showing technical feasibility for both cell localization and differentiation between band and segmented neutrophils, achieving a Mean Average Precision (mAP50) of 0.90. Furthermore, a web-based application prototype was developed as a proof of concept to demonstrate the model's inference capabilities. These findings suggest that Deep Learning technology has significant potential for automated cellular differentiation. In future developments, its direct integration into microscopy systems may serve as a valuable tool for supporting and standardizing left shift diagnosis.