A Hybrid Invasive Weed Optimization - Convolutional Neural Network and Bidirectional Generative Adversarial Network for White Blood Cell Image Segmentation and Classification in Leukemia Analysis

Accurate segmentation and classification of white blood cells (WBCs) are essential for automated hematological diagnostics, especially in early detection of leukemia. This study proposes a novel hybrid framework that integrates three complementary components: a U-Net-based segmentation model for precise extraction of WBCs from peripheral smear images, a Bidirectional Generative Adversarial Network (Bi-GAN) for synthetic data generation to address class imbalance, and a Convolutional Neural Network (CNN) classifier whose hyperparameters are optimized using the Invasive Weed Optimization (IWO) algorithm. Additionally, a fuzzy SoftMax layer is employed to enhance classification robustness in the presence of morphological ambiguity between WBC subtypes. The framework is evaluated on two widely used benchmark datasets, BCCD and LISC, achieving classification accuracies of ۹۹.۶% and ۹۹.۲۸%, respectively. Class-wise performance analysis using precision, recall, and F۱-score demonstrates the method's capability to effectively distinguish between all five WBC classes, including rare types such as basophils and eosinophils. The results confirm that the proposed system provides a reliable, interpretable, and computationally efficient solution for automated leukocyte classification and shows strong potential for deployment in real-time clinical settings.

کلیدواژه ها:

White blood cell classification ، Leukemia diagnosis ، U-Net Segmentation ، Bidirectional Generative Adversarial Network Augmentation ، invasive weed optimization ، convolutional neural network