Mamba-Based Ensemble Learning for White Blood Cell Classification
Lewis Clifton, Xin Tian, Duangdao Palasuwan, Phandee Watanaboonyongcharoen, Ponlapat Rojnuckarin, Nantheera Anantrasirichai · 2026
Accurate classification of white blood cells (WBCs) is critical for diagnosing hematologic and immunologic disorders. Traditional manual methods are labor-intensive and inconsistent, while existing automated systems such as the Sysmex DI-60 remain expensive and underperform on rare WBC subtypes. Deep learning has advanced this field, yet most approaches rely on computationally intensive Transformer-based models that are difficult to deploy in clinical settings. In this work, we propose an efficient and robust WBC classification framework based on Mamba, a recent state space model offering linear time complexity. We integrate five diverse Mamba variants into an ensemble architecture and introduce a new clinical dataset, Chula-WBC-8, comprising eight WBC classes with significant class imbalance. Extensive experiments on both Chula-WBC-8 and the BloodMNIST benchmark demonstrate that our approach outperforms state-of-the-art models and commercial analyzers. These results highlight Mamba's potential as a lightweight, accurate solution for real-world clinical workflows. The source code can be found at https://github.com/LewisClifton/Mamba-WBC-Classification.