Integrating convolutional neural networks for microscopic image analysis in acute lymphoblastic leukemia classification: A deep learning approach for enhanced diagnostic precision

Abstract

Leukemia is a type of cancer characterized by the exponential growth of abnormal blood cells, which damages white blood cells and disrupts the function of the human body’s bone marrow. It is very challenging to classify because blood smear images are complicated, and there is a lot of variation between each class. Acute Lymphoblastic Leukemia (B-ALL) is one of the subtypes of leukemia. It is a rapidly progressing cancer that originates in B lymphocytes, characterized by the overproduction of immature B lymphoblasts. The purpose of this work is to classify different types of B-ALL subtypes such as Benign, Malignant Early Pre-B, Malignant Pre-B, and Malignant Pro-B from the peripheral blood smear images effectively. To accomplish this task, a novel deep-learning technique based on a fine-tuned ResNet-50 model has been developed. Our fine-tuned ResNet-50 model integrates several additional customized fully connected layers, including dense and dropout layers. Various data augmentation techniques such as flipping, rotation, and zooming have been applied to mitigate the risk of overfitting. In addition, a five-fold cross-validation technique has been employed to enhance the model’s generalization. The performance of our proposed technique is compared with several other methods, including VGG-16, DenseNet-121, and EfficientNetB0, as well as existing baselines, using different performance metrics. Experimental results demonstrate the superiority of the fine-tuned ResNet-50 model, achieving the highest accuracy and an F1-score of 99.38%. It also outperforms existing state-of-the-art approaches by a significant margin. The proposed fine-tuned ReNet-50 model achieves such performance without the need for microscopic image segmentation which indicates its potential utility in healthcare sectors in enhancing precise leukemia diagnosis.