A Framework for Efficient Recognition and Classification of Acute Lymphoblastic Leukemia with a Novel Customized-KNN Classifier

Even in this modern era today, life's extent is still being challenged by many pathological diseases such as cancer. One such hazard is leukemia. Even a trivial setback in detecting leukemia lead to a severe outcome: the affected cells may eventually prove to be fatal. To combat this, we propose an algorithm to better segment the nucleus region of White Blood Cells (WBC) found in stained blood smear images with the intent of identifying Acute Lymphoblastic Leukemia (ALL). In our proposal, the image is made ready for segmentation in the preprocessing stage by changing its size, brightness, and contrast. In the segmentation stage, the nucleus region is segmented by mathematical operators and Otsu's thresholding. Then mathematical morphological operators are applied in post-processing stage, which makes the nucleus region convenient for feature extraction. Finally, the segmented regions are classified into ALL affected and regular cells by means of the proposed Customized K-Nearest Neighbor classifier algorithm. This work was experimented with over 80 images of the ALL-IDB2 dataset and attained an accuracy rate of 96.25%, 95% of sensitivity and 97% of specificity.