IMPROVING WHITE BLOOD CELL CLASSIFICATION WITH MINIMAL LABELS: A SEMI-SUPERVISED LEARNING FRAMEWORK

Abstract

The diagnosis of numerous blood disorders depends on the accurate classification of white blood cell
subtypes. Manually designed features are frequently needed for traditional computer vision techniques, which can be
time-consuming and performance-limiting. On the other hand, machine learning techniques provide higher accuracy
but usually require large labeled datasets, which are expensive and difficult to collect. This work presents a semisupervised learning strategy designed specifically for the classification of white blood cells. Through the utilization
of a limited quantity of labeled data and a more extensive collection of unlabeled data, the model is able to recognize
and classify various subtypes of white blood cells straight from microscopic pictures. This approach improves
classification performance by leveraging the data's natural structure and patterns rather than depending just on preestablished features.The suggested method was assessed on a dataset of artificial images that depicted different
subtypes of white blood cells. The results indicate promising accuracy in differentiating across cell types, indicating
potential uses in clinical diagnostics. White blood cell analysis in medical settings can be automated and made more
efficient with this scalable method that reduces the need for manually labeled data while retaining excellent
classification accuracy