Deep Learning-based Automatic Morphological Classification of Leukocytes using Blood Smears
Sumair Aziz, Muhammad Bilal, Muhammad Umar Khan, Fatima Amjad
Abstract
Classification of leukocytes into morphological classes is one of the main steps in the effective recognition of cancerous white blood cells and the diagnosis of malignancies. Before the evolution of computer-aided system designs, the process of morphological classification and inspection of microscopic blood smears is carried out by qualified and proficient human examiners which makes the procedure inefficient and slow. In this paper, a computer-aided classification technique is presented which is based on a combination of image processing and deep learning approaches to classify blood smears into its discrete morphological classes. The proposed methodology includes the conversion of blood images to L*a*b color space from RGB, K-mean clustering for segmentation, and deep convolutional networks for classification. Transfer learning is employed to shift the learned knowledge of pre-trained networks including AlexNet and ResNet18. Statistical evaluation parameters are used to validate the performance of the system on the Munich AML Morphological dataset. The results show reliable classification accuracy of 93.30% and 93.85% for AlexNet and ResNet18 respectively. The classification approach discussed in this paper can be used in real-time clinical environments as it is trained on a large dataset it allows detection in an earlier stage of the disease.