Enhanced Recognition of Acute Lymphoblastic Leukemia Cells in Microscopic Images based on Feature Reduction using Principle Component Analysis

Purpose- Acute lymphoblastic leukemia (ALL) is the most common form of pediatric cancer of white blood cells which is categorized into three types of L1, L2, and L3. It is usually detected through screening of blood and bone marrow smears by pathologists. Since manual detection is time-consuming and boring, computer-based systems are preferred for convenient detection. The rigorous similarity between morphology of ALL types and that of normal, reactive and atypical lymphocytes, makes the automatic recognition a challenging problem. In this paper, we tried to improve the sensitivity of detection based on principle component analysis (PCA). Methods- After segmenting cell nucleus using fuzzy c-means clustering algorithm, several geometric and statistical features are extracted. Then the feature space dimensionality is reduced based on (PCA). The first 8 components of the feature space are applied to support vector machine (SVM) classifier. Then the cancerous and lymphocyte cells are classified into their subtypes. Results- For evaluating the proposed method, we used an expert pathologist’s classification as a reference. Classification was evaluated by three parameters: sensitivity, specificity and accuracy. A comparison with our previous work showed that using dimensionality reduced feature space based on PCA, instead of using individually selected features, improved the average sensitivity and precision of classification more than 10%. Conclusion- The results show that proposed algorithm performs better than our previous work. Its acceptable performance for the diagnosis of ALL and its subtypes as well as other lymphocyte types makes it an assistant diagnostic tool for pathologists.