Klasifikasi Sel Darah Putih Menggunakan Vision Transformer (ViT)
Abstract
White blood cells are important for humans to protect the immune system from bacteria or viruses. White blood cells have 5 types of blood cells which consist of eosinophils, lymphocytes, monocytes, neutrophils, and basophils. Each type has a different shape. The classification of white blood cells is important in the field of healthcare to diagnose diseases caused by white blood cells, such as anemia, leukemia, and others. The classification process is usually done manually using a hemocytometer, the process is prone to human error which can lead to errors in diseases diagnosis. Advancement in technology, especially deep learning, can provide the potential to facilitate and minimize human error in the classification process. This report describes the application of a Vision Transformer (ViT) with a high-performance model in classifying white blood cells based on images. The ViT model uses multi-head attention to process information from images at the same time. The model training process uses a dataset containing 12,500 images and 4 types of white blood cells (eosinophils, lymphocytes, monocytes, neutrophils) with each having about 3,000 images. The model that has been trained using the right combination of hyperparameter values gets a train accuracy result of 98% and validation accuracy result of 83.44%. The model was used on a simple website as a test platform for classification. The results show that the model can classify white blood cells based on images correctly and can be a potential for the medical field.