“Detection of Acute Lymphoblastic Leukemia (ALL)” “Using Deep Learning and Transfer Learning”

Abstract

Leukemia results from abnormal and excessive multiplication of white blood cells, and as a result, the immune system in our bodies is destroyed and this leads to death. Annually, an estimated 300,000 new cases of Leukemia are diagnosed, which constitutes 2.8% of the total diagnosed cancers _[2]. Of the four different types of Leukemia, Acute Lymphoblastic Leukemia (ALL) is the most serious and deadly type of blood cancer. It spreads very quickly and is fatal within weeks if not treated. Statistically, about 90% of patients can be cured, provided the disease is diagnosed at an early stage _[2], so early diagnosis of (ALL) is definitely essential. In manual methods, pathologists diagnose it by a bone marrow biopsy test or an ocular microscopic testing of a blood sample. Although this method is very effective in diagnosis, it takes a long time and requires a lot of experience, so in this case, computer aided diagnosis (CAD) can be considered as a wonderful auxiliary diagnostic tool if computer aided diagnostic systems with high accuracy are developed to diagnose the disease that support Doctor's view. Several supervised and unsupervised machine learning algorithms have been proposed for the detection of (ALL).

In our work, we rely on samples of Acute Lymphocytic Leukemia micrographs from a dataset known as C-NMC being the largest available dataset aiming of building a deep learning model (a convolutional neural network CNN) and using the Transfer Learning for the EfficientNetB3 model to build a system that supports the Diagnostic view of the doctor. At the end of the work, after processing the data and training the model, we obtained a training accuracy of 98.623% and a testing accuracy of 97.75%, which is promising in the development of such systems.