White blood cells segmentation and classification using U-Net CNN and hand-crafted features
Paper ID: 1938
Hematological disorders can be diagnosed based on the white blood cells (WBCs) count. However, determining WBCs count manually from a microscopic blood smear image is a tedious task and can be done through automatic detection and classification of WBCs using image processing techniques. In this paper, an efficient WBCs detection and classification method is presented. Towards ameliorating this, U-Net CNN segmentation architecture is utilized to detect and segment out the WBCs from red blood cells and platelets. The segmented WBCs are classified into five types based on their texture and shape features using an SVM classifier. Results are presented for the LISC dataset using a 5-fold cross-validation scheme. The performance measures indicate that the proposed method can perform better than similar techniques.