ABSTRACT
The human immune system plays a critical role in defending against
infections and diseases, with white blood cells (WBCs) being pivotal
in these processes. Automated classification of agranulocyte
cells, specifically lymphocytes, and monocytes, is essential for accurate
diagnostics and treatment monitoring in hematology and
oncology. This study evaluates the performance of a convolutional
neural network (CNN) model, previously proposed for WBC classification,
on public datasets, with and without the use of Contrast
Limited Adaptive Histogram Equalization (CLAHE) for image preprocessing.
The results show that CLAHE improved classification
metrics, achieving up to 82.16% test accuracy on the Paul Mooney
dataset and maintaining a high test accuracy of 98.72% on the Uncle
Samulus dataset. Metrics such as precision, recall, and F1-score also
exhibited notable improvements, reaching up to 98% for lymphocytes
and monocytes in the best-performing dataset. These findings
highlight CLAHE’s potential to enhance CNN-based classification
under varying image conditions.