Classification of normal and leukemic blast cells in B-ALL cancer using a combination of convolutional and recurrent neural networks

Abstract

Acute Lymphocytic or Lymphoblastic Leukemia (ALL) is a virulent form of blood cancer that affects white blood cells and the bone-marrow–spongy tissue. At the start of ALL, immature white blood cells proliferate and replace healthy cells in the bone marrow. ALL progresses quickly and can be fatal within a few months if not treated. Computer assisted diagnosis and prognosis of ALL, therefore, has the potential to save many lives but requires high accuracy classification of malignant cells which is challenging due to the visual similarity between normal and malignant cells. In this work, we employ a custom-built deep learning model for the classification of immature lymphoblasts and normal cells. Our model is an ensemble of convolutional and recurrent neural networks. It also exploits the spectral features of the cells by using discrete cosine transform in conjunction with an RNN. The proposed classifier has been validated using multiple experiments. Our approach is able to achieve substantial performance gains when compared to, conventional, stand-alone CNN- and RNN-based methods. The highest accuracy achieved by our model is 86.6%.