Fuzzy logic and deep learning approach for automated white blood cell detection and classification via multi-CNN feature fusion.

White blood cells (WBCs) are vital components of the human immune system, responsible for defending the body against pathogens and infections. The classification and detection of white blood cells (WBC) are indispensable for the diagnosis of hematologic diseases such as leukemia, but manual classification is time-consuming and susceptible to subjectivity. This paper introduces a new framework that integrates multi-CNN feature fusion with fuzzy logic-based Evaluation Based on Distance from Average Solution (EDAS) for automatic classification and detection of WBCs. The method combines aspects of DenseNet121, MobileNetV2, and ResNet101 to obtain varied morphological features, tackling intraclass variation and imaging inhomogeneity. A fuzzy EDAS model prioritizes the fused features for better interpretability and robustness in clinical decision-making. Using the Kaggle Blood Cell Images dataset of 8013 images of neutrophils, eosinophils, monocytes and lymphocytes, the suggested approach obtained an overall precision of 99.79%, precision, sensitivity and F1 scores greater than 99.70% for all types of WBC. In contrast to individual CNN models (see, e.g., DenseNet121: 95.28%, ResNet101: 81.85%), the combined model performs much better, especially for difficult classes such as neutrophils. The fuzzy EDAS method also guarantees robust model classification in an uncertain situation and is therefore appropriate for clinical settings. It minimizes diagnostic delay, improves scalability in low-resource scenarios, and facilitates quick screening for blood diseases. Future development will address real-time implementation and verification on varied datasets to ensure generalizability.