AUTOMATED BLOOD CANCER CLASSIFICATION USING DEEP NEURAL NETWORKS

Abstract

Improving survival rates for blood cancer, particularly leukemia, depends on early detection. Traditional diagnostic techniques depend on the labor-intensive and subjectively interpretable manual microscopic examination of peripheral blood smear pictures. Deep learning advances in recent years have made automatic detection more accurate and reliable. A labeled leukaemia dataset with four classes—Benign, Early, Pre (Pre-B), and Pro (Pro-B)— is used in this study to train and assess the model. Deep CNN designs like VGG16 and VGG19, which offer powerful feature extraction but are computationally demanding, were mostly used in earlier studies. MobileNetV2, a portable and effective deep learning model tailored for real-time diagnosis, is proposed in this work. For reliable classification, the suggested approach combines deep feature learning, image preprocessing, and augmentation. The results of the experiments show that the MobileNetV2 based model improves diagnosis speed while retaining excellent accuracy, which qualifies it for practical clinical implementation