Automated Acute Lymphocytic Leukemia (ALL) Detection Using Microscopic Images: An Efficient CAD Approach

Abstract

Leukemia, which is caused by the excessive and aberrant reproduction of white blood cells, completely destroys the immune system of our body and leads to death. Among four different types of leukemia, the progress of acute lymphocytic leukemia is rapid and becomes fatal even in weeks if it is kept untreated. So, early diagnosis of ALL is quite necessary. In manual methods, pathologists diagnose ALL by the microscopic test of blood specimen or bone marrow test. Though this is the most efficient process to diagnose ALL, it is a time-consuming matter. In this case, computer-aided diagnosis (CAD) may be considered as a great associative diagnostic tool for ALL identification. Numerous supervised and unsupervised machine learning algorithms have been proposed for ALL detection for years. This paper concerns with establishing a CNN- based CAD system for automated ALL detection from the microscopic blood images which is collected from ALL-IDB dataset. In this regard, at first the images have been preprocessed applying median filter and histogram equalization for the purpose of reducing the noise and enhancing the image. Being smaller in size, data augmentation has been applied on the dataset which increases the images by including slightly modified copies of images that already exist in the dataset. Finally, the modified images are passed through a CNN model for training purpose where feature extraction and classification are performed by convolution, ReLU, pooling layer, and a fully connected layer. Here, the dataset is also trained using some pretrained model to show a comparison of our model with other models. It is observed that our proposed model results as a well fitting model with 100% training accuracy and 97.89% testing accuracy which is promising.