Automatic Recognition of Acute Lymphoblastic Leukemia using Multi-SVM Classifier

Acute lymphoblastic leukemia (ALL) is the most popular form of white blood cells cancer in children. It is classified into three forms of L1, L2 and L3. Typically, it is identified through screening of blood smearsvia pathologists. Since this is laborious and tedious, automatic systems are desired for suitable detection; but the high similarity between morphology of ALL forms and that of normal, reactive and a typical lymphocytes, makes the automatic detection a challenging problem. This study tried to improve the accuracy of detection based on principle component analysis (PCA). After segmenting nuclei of cells, numerous features were extracted. The first six components of this feature space were used for the binary and multiclass support vector machine classifiers. An expert pathologist was used to appraise this method as a gold standard. A collation with similar work indicated that using PCA instead of using exclusively selected features enhanced the average sensitivity and specificity of classification up to 10%. The results demonstrate that this algorithm performs better than similar studies. Its permissible efficiency for identifying ALL and its sub-types as well as other lymphocyte forms makes it an associate diagnostic device for pathologists.