ACUTE LYMPHOBLASTIC LEUKEMIA SUBTYPE CLASSIFICATION USING CONVOLUTIONAL NEURAL NETWORKS ON PERIPHERAL BLOOD SMEAR IMAGE

Abstract

Acute Lymphoblastic Leukemia (ALL) is the most common kind of cancer in children. It is caused by too many immature lymphoblasts growing in the bone marrow. Accurate subtype identification is crucial for timely and effective treatment. This work presents a deep learning approach for the automated classification of four diagnostic categories based on microscopic peripheral blood smear images: benign (hematogones) and three malignant subtypes of acute lymphoblastic leukemia (Early Pre-B, Pre-B, and Pro-B). Transfer learning was used to improve four pre-trained convolutional neural network architectures EfficientNet-B3, VGG16, DenseNet-121, and ResNet-50—on a dataset of 3,256 images. EfficientNet-B3 achieved the highest test accuracy of 98.57%, followed by VGG16 (98.37%), DenseNet-121 (97.76%), and ResNet-50 (95.92%). The proposed strategy demonstrates enhanced diagnostic precision and has considerable potential to reduce observer variability, minimize diagnostic errors, and expedite clinical decision-making in all screening and subtype identification processes.

Keywords: Acute Lymphoblastic Leukemia, Convolutional Neural Networks, Blood Smear Classification, Deep Learning, Automated Diagnosis