Hybrid Deep Learning Framework for White Blood Cell Classification Using Attention-Based Feature Fusion
Samah Adel Gamel, Mervat El-Seddek · 2025
Accurate classification of white blood cells (WBCs) plays a vital role in medical diagnosis and treatment planning for various hematological disorders. Manual identification is often time-consuming, error-prone, and requires expert knowledge. In this paper, we propose a novel hybrid deep learning framework that leverages transfer learning and attention-based feature fusion for automated WBC classification. The proposed model combines feature representations from two powerful pre-trained convolutional neural networks-ResNet50 and Mo-bileNetV2-followed by a self-attention mechanism to enhance feature discriminability. The final classification is performed using fully connected layers with softmax activation. Experimental results on a publicly available, augmented WBC dataset comprising 12,500 labeled images across four WBC types-Eosinophil, Lymphocyte, Monocyte, and Neutrophil-demonstrate the effectiveness of the proposed approach, achieving a classification accuracy of 98.5%. The robustness and interpretability of the model are further validated using standard evaluation metrics and Grad-CAM visualizations. This work demonstrates the effectiveness of combining multiple pretrained CNNs and attentionbased fusion for WBC classification, offering a balance between accuracy and interpretability for real-world diagnostic systems.