Abstract Acute Lymphoblastic Leukemia (ALL) is a rapidly developing blood cancer that requires early detection for effective treatment. Manual identification of leukemic cells from microscopic blood smear images requires a lot of time and can be easily mistaken. This research introduces a hybrid deep learning framework that integrates U-Net for segmentation and Vision Transformer (ViT) for classification using the ALL-IDB2 dataset. The approach involves data preprocessing, augmentation, and region segmentation to isolate leukemic cells. During the classification stage, optimizers such as Adam, Adamax, and Nadam are used to enhance model convergence and improve classification accuracy. Experimental analysis reveals that the suggested model achieves an impressive accuracy of 98.7%, surpassing traditional CNN-based & Transformer based Architectures. The study validates the efficacy of transformer-based hybrid models in medical image analysis and establishes a reliable framework for automated leukemia diagnosis.
Revathi et al. (Fri,) studied this question.