Volume - 7 | Issue - 2 | june 2025
Published
04 July, 2025
Proper classification of white blood cells (WBCs) remains a critical task for medical diagnosis in leukemia, infections, and hematological disorders. WBC classification is based mainly on Convolutional Neural Networks (CNNs) as the main methodology, despite the challenges in such methods in scanning microscopic images for distant dependencies. The current study adopts a hybrid model that combines Vision Transformers (ViTs) with CNNs for better WBC classification. The model has a ViT component that uses self-attention mechanisms to obtain whole-image features through global features extraction, while at the same time, neighborhood information is being extracted by the CNN module. The introduced model is evaluated using theBCCD Dataset (Blood Cell Count and Detection) for binary and multi-class separation of granulocytes/agranulocytes. Experimental tests indicate that the combined approach has robust classification accuracy with guaranteed reliable results in various evaluation metrics. The suggested method achieved 99.20% training accuracy, 87.90% test accuracy, a precision of 0.7083, a recall of 0.7000, and an F1-score of 0.6970, affirming its excellent classification precision. This hybrid deep learning model shows enhanced performance and interpretability, supporting its application in clinical diagnostic processes. These findings attest to excellent classification performance, positioning the model as an excellent choice for automated hematological diagnosis. The study opens up new horizons for combining CNNs and ViTs for improved medical image analysis while setting future development goals in WBC classification. Deep learning hybrid models have shown their significance for clinical diagnostics in the study through the development of enhanced detection systems against hematological diseases.
KeywordsVision Transformers White Blood Cells Hybrid Model F1-Score Deep Learning
