CLAHE enhanced hybrid feature descriptors for classification of acute lymphoblastic leukaemia in blood smear images

Acute lymphoblastic leukaemia (ALL) detection through a complete blood count test is often flagged to an expert pathologist for confirmation which is time-consuming, observer-specific and involves intensive labour. The study proposes an efficient computer aided diagnosis (CAD) method based on image processing and machine learning models to assist doctors in analysing microscopic images. This study aimed to investigate the combined discriminative qualities of shape and texture features, as well as the best fit feature subset selection technique, to achieve high accuracy and a low false positive rate for classification of healthy and ALL infected leukocyte cell images. This approach outperformed existing models with an accuracy of 92.3%, a precision of 96%, and a false positive rate of 3.846%. As a result, the proposed methodology is capable of more precisely classifying the images into healthy and ALL affected cell images, assisting physicians in the detection and diagnosis processes.