A Transfer Learning-Based Deep CNN Approach for Classification and Diagnosis of Acute Lymphocytic Leukemia Cells
Leo Dominick C. Magpantay, Helcy D. Alon, Yolanda D. Austria, Mark P. Melegrito, Glenn John O. Fernando
Abstract
Leukemia is a severe malignant illness affecting both adults and children. It can be chronic or acute. Acute Lymphoblastic Leukemia (ALL) is the most prevalent type of blood disease and one of the leading causes of death. This type of cancer invades the blood and spread through neighboring organs and body systems. To classify cancer cells and non-cancer cells, specialists need to perform manual diagnosis through inspection of cell images under a microscope and provide labels through annotation. However, this manual microscopic analysis is tedious and may lead to a false diagnosis. With this concern, the author proposed a computer-aided diagnosis method using a transfer learning-based deep learning approach. In this study, the YOLOv3 model is utilized to train a deep learning model that will classify ALL and normal cells. The model produced promising results as it has a training loss of 2.8% or a 97.2% training accuracy. Based on the model evaluation, the model has an mAP value of 99.8% as well. YOLOv3 was shown to be a useful tool for distinguishing leukemia cells from non-leukemia cells.