Detection and Classification of Acute Lymphoblastic Leukemia using CNN

Acute lymphoblastic leukemia is a very important cancer in childhood but quite prominent in later years of life for the genetic defects in lymphoid progenitors, which are hallmarks of the disease. In children, ALL mostly affects those aged between 2 and 6 years old and, against the background of contemporary biology knowledge and treatment approaches, is associated with more than 80% cure rates. However, approximately 20% of children with ALL relapse; therefore, there is a huge need for better risk identification and treatment optimization. In adults, ALL mainly affects B-cell precursors and is treated with chemotherapy and, in some cases, stem cell transplantation. An accurate and early diagnosis of ALL is of key importance but difficult to realize due to morphological similarities between normal cells and leukemic cells. This study, therefore, proposes a CNN model to improve diagnostic accuracy. Furthermore, it exploits the capabilities of CNN in feature extraction with Adamax Optimizer and the Categorical Cross-Entropy Loss Function to deal with imbalances and noise in the dataset. RESNET50-CNN has achieved 98.63% accuracy in classification and is hence a very strong tool in ALL detection and classification.