SEPO-WBC: Automated white blood cell analysis software using deep learning for veterinary hematology

초록

The analysis of blood smear images for the classification of white blood cell (WBC) subtypes, known as the differential count, is fundamental to hematological diagnosis. However, conventional manual analysis is timeconsuming, labor-intensive, and prone to inter-observer variability. To address these challenges, this study presents SEPO-WBC (Software-based Evaluation Process for Optimized White Blood Cell analysis). This open-source software automates WBC detection and classification in canine blood smear images to calculate differential counts. A two-stage pipeline is employed, utilizing the You Only Look Once (YOLO) model for detection and the Densely Connected Convolutional Networks (DenseNet) model for classifying detected WBC subtypes, with each model selected for optimal performance in its respective task. Experimental results demonstrate that SEPO-WBC achieves an mAP50 value of 0.935 for detection and an accuracy of 0.878 for classification. A clinical comparison with five experts confirms detection equivalence with an intraclass correlation coefficient (ICC) of 0.969 and clas sification agreement of 0.855-0.883 within an inter-expert range of 0.845-0.970, while also realizing much faster processing.

키워드