Diagnosis of Leukemia Disease through Deep Learning using Microscopic Images

Abstract

The abnormal production of WBC (white blood cell) in the bone marrow is known as leukemia. Leukemia is one of the most affecting diseases around the globe. Several types of ML (machine learning) and DL (deep learning) classification models have been presented in the literature to detect this disease, but they still possess some drawbacks. This study proposes a framework for detecting five classes of leukemia: ALL, AML, CLL, CML, and normal cell. In this research, a pre-trained DCNN (deep convolutional neural network) has been proposed for the detection of leukemia through microscopic images. Pre-processing of microscopic images improves the contrast and removes the noise by enhancing and filtering images. Segmentation of microscopic images is used to highlight the area of the disease. Alex-Net and ResNet-34 architecture are used for classification purposes. After comparing these two models through statistical parameters, ResNet-34 attained the most accurate result than Alex-Net using the publicly available ALL-IDB dataset, the evaluation through the statistical parameters revealed that ResNet-34 attained a classification accuracy of 98.4% on ALL, 98.4% on AML, 98.13% over CLL, 98.14 over CML. AlexNet attained 96.1% classification accuracy on ALL, 95.5% on AML, 95.7% on CLL, and 96.8% on CML. The proposed framework significantly outperforms existing technologies and can be used in clinical applications.