An Efficient Detection and Classification of Acute Lymphoblastic Leukemia using CNN models

One of the major ubiquitous kinds of leukemia is ALL (Acute lymphoblastic leukemia) and potentially lethal hematological malignancy is distinguished by the uncontrollable proliferation of premature lymphocytes in peripheral blood and bone marrow stem. Traditional Acute lymphoblastic leukemia diagnosis, which is done by competent examiners employing microscopic images of a smear of the peripheral blood, is labor-intensive and time-consuming. CNN(Convolutional Neural Networks) is currently the prioritized option for histopathology image processing. Conventional CNN typically requires massive databases for adequate training in order to achieve excellent performance. This paper suggests a prompt, intense CNN architecture to address which concern and achieve better identification of ALL. A unique probability-based parameter is proposed, which has a substantial impact in dexterously hybridizing Vgg16, GoogleNet, and AlexNet while preserving the advantages of each individual approach. Further, the models are trained and validated with various parameters, the algorithms with the best parameters were applied to the test set. Among models, GoogleNet, AlexNet, and VGG-16 achieved 97%, 96.52%, and 98% accuracy, respectively. The model also shows high precision and recall values for both healthy and leukemia-affected WBC. The result shows that the proposed method could indeed aid in the diagnostic test of ALL by inspecting immature leukocytes efficiently.