DAFFNet: A dual attention feature fusion network for classification of white blood cells

• We proposed a novel dual-branch Dual Attention Feature Fusion Network, merging morphological features with high-level semantic features for WBC classification. • Combining EPSA and SA, we introduced dual attention mechanism, enabling the model to comprehend and utilize crucial features comprehensively. • We proposed MFE, consisting of two modules: MAP for extracting morphological attributes and MAE for encoding them, enhancing the model’s classification capability. • Utilizing category labels and unlabeled samples, we trained MAP using DSL and SSL, improving its performance in morphological attribute extraction. • We employed six datasets to verify the model’s robustness and effectiveness across multiple center, introducing Labelled and WBCAtt datasets for the first time. The precise categorization of white blood cells (WBCs) is vital for diagnosing blood-related disorders. However, manual analysis in clinical settings is often time-consuming, labor-intensive, and prone to errors. Consequently, the use of Computer-Aided Diagnostic (CAD) techniques has become essential to assist hematologists in accurately classifying WBCs, improving diagnostic efficiency and reliability. Numerous studies have employed machine learning and deep learning techniques to achieve objective WBC classification, yet these studies have not fully utilized information on WBC images. Therefore, our motivation is to comprehensively use the morphological and high-level semantic information of WBC images to achieve accurate classification of WBC. In this study, we propose a novel dual-branch network, the Dual Attention Feature Fusion Network (DAFFNet), which first integrates the high-level semantic features with the morphological features of WBC to achieve more accurate classification. Specifically, we introduce a dual attention mechanism, enabling the model to comprehensively leverage both the channel-wise and spatially localized features of the image, enhancing its ability to capture crucial information for precise WBC classification. A Morphological Feature Extractor (MFE), comprising Morphological Attributes Predictor (MAP) and Morphological Attributes Encoder (MAE), is proposed to extract the morphological features of WBC. We also implement Deep-supervised Learning (DSL) and Semi-supervised Learning (SSL) training strategies for MAE to enhance its performance. Our proposed network framework achieves 98.77%, 91.30%, 98.36%, 99.71%, 98.45%, and 98.85% overall accuracy on the six public datasets PBC, LISC, Raabin-WBC, BCCD, LDWBC, and Labelled, respectively, demonstrating superior effectiveness compared to existing studies. On the BCCD dataset and Labelled dataset, the overall accuracy of our model exceeds the state-of-the-art model by 0.52% and 4.36%, respectively. The results indicate that the WBC classification combining high-level semantic features and low-level morphological features is of great significance, which lays the foundation for objective and accurate classification of WBC in microscopic blood cell images.