A Deep Learning Technique for Multi-Classification of Acute Lymphocytic Leukemia
Mervat El-Seddek · 2024
Early detection, diagnosis, prognosis, and treatment of blood cancers are challenging healthcare problems. Leukemia is a very fatal blood cancer. Leukemia can be categorized into acute leukemia and chronic leukemia. In childhood, acute lymphocytic leukemia (ALL) is responsible for about 25% of all childhood cancers. Because the microscopic images of acute leukemic b-lymphoblasts are very similar to those of benign cells under the microscope, it is very hard to stage cancer and differentiate them from benign cells. The present work proposes a computer-aided detection system to classify ALL into four classes, based on convolutional neural network (CNN) models, namely VGG16 and DenseNet201. These classes include normal cells, early lymphoid precursor (ELP), pro-B cells, and pre-B cells. The classification accuracy of the suggested approach had reached 99.66% and 99.9% for ALL multi-classification using VGG-16 and DensNet-201 respectively. The suggested system potentially improves the accuracy and speed of diagnosis to help pathologists and specialists analyze images and make the most appropriate decision to identify clinical cases of patients and start an early treatment approach.