Automated Classification of Leukocytes: Deep Learning Model Analysis Based on Granule Content
Rabiye Kılıç, Hilal Kübra Sağlam, Emin Argun Oral, İbrahim Yücel Özbek · 2025
Automatic classification of white blood cells according to their granule content is of critical importance for hematological analysis and disease diagnosis. In this study, the performances of DarkNet-19 and DarkNet-53 models were evaluated for deep learning-based classification. First, two-class classification was performed for the separation of Granulocytes and AnGranulocytes, and the DarkNet-19 model achieved 94.8% and DarkNet-53 model 98.0% accuracy. Then, each cell type was analyzed separately with six-class classification, and DarkNet-19 achieved 91.37% and DarkNet-53 achieved 96.53% accuracy. In particular, the DarkNet-53 model showed a significant improvement by providing higher F1-scores in cell types with low samples. The findings show that the DarkNet-53 model can distinguish cell morphology more successfully thanks to its deeper architecture and is a strong candidate for integration into automatic leukocyte classification systems.