Deep Learning-Driven Classification of Acute Lymphoblastic Leukemia Subtypes Using CNNs

In this paper, a Convolutional Neural Network (CNN) based model is created to differentiate blood cell images into the benign and malignancy classes with more emphasis on subtype of acute lymphoblastic leukemia (ALL). The model is trained on a dataset of 3,242 high-resolution images from 89 patients with annotations prepared by seasoned medical practitioners. The CNN successfully distinguished between four classes. The identification of the leukemias is done according to the WHO categorization; they include Benign, Malignant Pre-B, Malignant Pro-B and Malignant early Pre-B and the accuracy obtained is 98.3%. This high level of accuracy is supported by the high precision, recall, and F1-scores of the model for all the classes that are tested. According to the findings of the research, deep learning and specifically CNNs can be effectively used to enhance the diagnostic performance and reduce the diagnostic time for blood cell cancer. They could also improve diagnostic procedures in clinical practices, thus improving overall patient care and timely interferences. This work can therefore be used as a model to show how AI tools can be incorporated into the daily practice of medicine, and provides a sound and practical solution to hematological testing.