Deep Learning Methods for Chronic Myeloid Leukaemia Diagnosis

Abstract

AbstractChronic myeloid leukaemia (CML) is considered as the cancer of the blood or cancer of the bone marrow. Blood consists of three different cells, one that helps to fight diseases and some infections called white blood cells (WBCs), another that can carry oxygen to the tissues of the body called red blood cells (RBCs), and, last but not the least, those which help the formation of blood clots to stop bleeding termed as platelets. Cancer is observed as a disease ‘diversified in character’, which consists of many different subtypes. Blood cancer is the most common toxicant for the human body.It is said to arise when the development of blood cells stops them from working properly. Chronic myeloid leukaemia is regarded as a myeloproliferative disease which is distinguished by the viable presence of Philadelphia chromosome, also known as Ph chromosome correspondingly associated with the BCR-ABL fusion gene. The proportion of patients are asymptomatic as the diagnosis is done after the disease is all through in the entire body. Therefore, the detection of CML requires deep machine learning. The study proposes new approach for diagnosing CML which requires a large training dataset. The significant available CML data source used here is BioGPS, from where the training dataset was collected so as to explore machine learning algorithms. The different machine learning algorithms used here are as naïve Bayes classifier, k-nearest neighbour algorithm, support vector machine algorithm, decision tree method, etc. Deep learning has been used taking into consideration artificial neural networks (ANN). From the variants of ANN, the convolutional neural network (CNN) is the most computational model. An architecture has been designed for deep learning which uses the concept of CNN for detection, extraction, segmentation, and classification of different phases of CML and also the presence of CML identification, which further can help the physicians to identify the present state of the patient for the appropriate treatment. The implementation of the algorithms is done to improve some accuracy and for better performances as compared with the previous forms of artificial intelligence among diverse machine learning methods.KeywordsChronic myeloid leukaemiaArtificial neural networksConvolutional neural networksDeep learning algorithmsNaïve Bayes classifier