Vision Transformer (ViT) approach in computer aided diagnosis of acute lymphoblastic leukemia

Abstract

Blood cancer is a serious and potentially deadly type of cancer that affects the pro duction of blood cells in the body.Leukemia, lymphoma, and myeloma are the three primary kinds of blood cancer.Leukemia, which is the most common and deadly type of blood cancer, is characterized by the abnormal and unexpected development of white blood cells (leukocytes) in the bone marrow. Leukemia comes in two primary varieties: acute and chronic. Acute leukemia progresses more quickly and is more common in children, while chronic leukemia progresses more slowly. Early detec tion of leukemia is important for proper treatment, as it can be fatal if not treated promptly. One method of detecting leukemia is through imaging, which is quick and inexpensive and does not require specialized equipment or laboratory tests. How ever, manual classification of leukemia cells by hematologists can be time-consuming and prone to errors. In recent years, the preferred technique for vision application is convolutional neural networks (CNNs).CNN have demonstrated their effectiveness in automatically classifying medical images. However, their limited local receptive field can prevent them from learning global context information. An alternative to CNNs that has shown promise is the Vision Transformer (ViT), which uses self-attention between image patches to process visual information. However , Vit does not work very well without a large dataset so we are using the ISBI 2019 data set, a dataset of 10000+ images and this data set needs more polishing, we’re not just suggesting a transformer architecture for diagnosing ALL; we’re also laying the groundwork for its polishing and sharing every piece of code we’ve used in our research. Our Vit model produces an accuracy of 81.5%, and shows how it has potential to reach new heights. The suggested approach has the ability to accurately differentiate between cancer cells knows as B-lymphoblast cells and normal cell known as B-lymphoid precursors and can be utilized as an efficient technique for assisting in the effective discovery of acute lymphoblastic leukemia through computer assistance.