Generative-Augmented Deep Feature Representation Learning for Accurate Blood Cancer Diagnosis from Histopathological Images

Background: Leukemia is a cancer that impacts the blood and bone marrow. Detection and classification are traditionally performed using labor-intensive and specialized techniques. Detecting blood cancer through histopathological analysis of bone marrow tissue is a critical diagnostic process that requires accuracy and efficiency. In this study, we propose an automated system for detecting blood cancer (Leukemia) from bone marrow histopathological images using a hybrid approach combining Generative Adversarial Networks and Convolutional Neural Networks (GAN+CNN). Methods: The system processes high-resolution images to identify malignancies by learning complex patterns from pixel-level features. Our model incorporates image preprocessing techniques such as stain normalization and noise reduction to improve image consistency and quality. This automated approach can speed diagnosis and reduce subjectivity in blood cancer detection, providing a robust tool to support clinical decision-making. Results: The proposed GAN+CNN achieves 99.96% accuracy, 98.3% precision, 98.4% recall, and a 97.53% F1-score. Conclusion: The proposed GAN+CNN method achieved better performance in leukemia detection and can support rapid, reliable diagnosis.