Identification of different types of lymphoblasts in acute lymphoblastic leukemia using relevance vector machines

Abstract

Acute Lymphoblastic Leukemia (ALL) is a blood cancer, which is characterized by an abnormal proliferation of lymphoblasts (a form of white blood cells). It leads to debilitating condition of the patient and necessitates an early therapeutic intervention. For diagnosis of ALL, a pathologist examines bone marrow smear for the count of different types of lymphoblasts to arrive at the diagnosis. The different counts of three different lymphoblasts can affect the type and intensity of the therapy to be used as well as the likely course of the disease. However, bone marrow smear examination and its interpretation is a tedious time consuming process. Therefore, any automation in terms of easing and pacing up the workflow is required. In this paper, we exactly do that by proposing an automatic method for recognition of different types of cells in ALL. The proposed algorithm mimics the pathological and ontological descriptions of bone marrow cells, in identifying abnormalities that lead to ALL. The proposed method uses discriminative shape, colour and texture features, which supposedly contains information for better discrimination of bone marrow cells. Further feature selection techniques, based on mutual information distribution and recursive feature elimination along with Relevance Vector Machines (RVM) are used for effective classification. The results are analyzed on more than 345 cell images. Sensitivity and specificity of above 93% has been achieved.

A convolutional neural network–based learning approach to acute lymphoblastic leukaemia detection with automated feature extraction

“…Marker control watershed segmentation was also used in some other notable contributions for ALL classification. This was able to detect attributes of WBCs such as color and shape [14][15][16][17]. A threshold-based method to perform white blood cell segmentation from microscopic image was also proposed in [18].…”