Article Preview
TopIntroduction
Leukemia (blood cancer) is a disease that forms in the tissues responsible for the production of blood cells, which include bone marrow and the lymphatic system. This type of cancer appears firstly within white blood cells. The latter is responsible for bodies’ safety against various infections. These cells generally grow naturally and are divided according to the needs of the body. However, in the case of leukemia, the bone marrow produces a very large amount of abnormal white blood cells, which cannot function properly. Treatment for leukemia should be achieved as earlier as possible to get better results for patients. When doctors notice signs or symptoms that suggest leukemia, patients may undergo diagnostic exams such as physical blood and bone marrow test. Doctors may also involve flow cytometry technique (Marchisio et al., 2021) to confirm the presence of leukemia cells. Although flow cytometry is the best standard for accurate and automated measurement of malignant white blood cells, it is very expensive and requires sophisticated equipment as well as trained personnel to perform it. In addition, the lack of ready access to technical support and quality assurance programs has limited the use of flow cytometry techniques, especially in developing countries.
In the last few years, new techniques based on artificial intelligence and machine teaching (Douglass, 2020) have achieved great success in many fields, especially in the health sector and medical images classification. This new technology is easier and cheaper than most standard techniques. It helped doctors and medical laboratories to solve many medical problems and assisted in diagnostics. Machine learning is an artificial intelligence technology that gives computers the ability to learn without being explicitly programmed. Among the better definition of machine learning that is presented in (Douglass, 2020): “A machine learning model is a trained computer program based on experience E concerning some new tasks T and some performance measures P. Its performance on T, as measured by P, should improve with experience E”. Deep learning is a subset of machine learning based on artificial neural networks. It is considered an evolution of machine learning that enables machines to make accurate decisions without human help.
Many deep learning algorithms have proven their efficiency and success in medical images classification. They help doctors to make quick decisions about image samples if they are normal or abnormal. In the case of severe diseases, it is often difficult to distinguish between normal and abnormal elements (cells, organs, etc.) due to their similar morphology. For example, in the case of leukemia, a deep learning model assists doctors to decide whether white blood cells are malignant or healthy, based on microscopic images. Among the most widespread deep learning algorithms is Convolutional Neural Networks (CNN or ConvNet) (Douglass, 2020). It is a special architecture of artificial neural networks, most commonly applied to analyze and classify visual imagery (Douglass, 2020). CNN is a multilayered algorithm where the outer layers are dedicated to extracting features from images, while the inner layers are a neural network intended for features classification.
In this work, we propose a CNN model for the classification of white blood cells from microscopic images. To reach this goal we have used the dataset published by CANCER IMAGING ARCHIVE (Natasha, 2021). This dataset is composed of malignant white blood cell images and normal (healthy) white blood cell images.