Publication

Abstract : Leukemia, a highly malignant form of blood and bone marrow cancer, can be identified through meticulous examination of blood smears under a microscope. This research is to enhancing the early detection of acute lymphoblastic leukemia (ALL) in patients, thereby enabling prompt and life-saving interventions. The distinctive challenge in identifying ALL lies in the shared traits between lymphocytes and lymphoblasts, notably in the nucleus segmentation stage, a pivotal component of diagnosis. In response to this challenge, we introduce an innovative strategy that employs a CNN-based U-net segmentation model. Our model integrates a pre-trained hybrid feature extractor, seamlessly combining attributes from the VGG19, InceptionResnetV2, and MobilenetV2 architectures. Performance evaluation involves utilizing the openly accessible ALL-IDB2 dataset and employing essential metrics, including the dice coefficient, Jaccard index, and accuracy. Significantly, our approach not only outperforms current state-of-the-art methods such as VGG16, VGG19, Alex Net, Google Net, Squeeze Net, and Xception, but it also achieves a notable 98.4% accuracy rate and an exceptional dice score of 98.9%, underscoring its remarkable superiority.