Domain Knowledge Incorporation in WBC Classification

White Blood Cells (WBCs) play a vital role in the human immune system, helping in the diagnosis of various illnesses and monitoring patient health. This project was conducted to explore different approaches for the classification of WBCs from blood smear images, wherein domain knowledge is included in the deep learning model. As traditional approaches for WBC classification frequently rely on subjective and time-consuming manual assessment with or without computer-aided diagnosis (CAD), In order to improve accuracy, this study highlights a novel deep learning technique that combines neural networks and domain knowledge. By leveraging expert insights on haematological characteristics, the developed model generates a knowledge vector. This generated knowledge vector is further used for classification in two different approaches, the first one being Graph Traversal and the second being Decision Fusion. As a result, the image is classified into four specific WBC types, which are Neutrophils, Eosinophils, Monocytes, and Lymphocytes. Further, to ensure that the model encounters a diverse range of WBCs belonging to all classes, a strategic sampling technique has been employed before training. This interpretable method has automated and streamlined WBC classification, and it is hoped to reduce the workload on medical staff, making it easier to recognise early disease, with the change in normal appearance of a WBC for better patient care.