Vol.:(0123456789) 1 3 Australasian Physical & Engineering Sciences in Medicine https://doi.org/10.1007/s13246-019-00742-9 TECHNICAL PAPER Feature extraction using traditional image processing and convolutional neural network methods to classify white blood cells: a study Roopa B. Hegde 1,2  · Keerthana Prasad 1  · Harishchandra Hebbar 1  · Brij Mohan Kumar Singh 3 Received: 19 March 2018 / Accepted: 25 February 2019 © Australasian College of Physical Scientists and Engineers in Medicine 2019 Abstract White blood cells play a vital role in monitoring health condition of a person. Change in count and/or appearance of these cells indicate hematological disorders. Manual microscopic evaluation of white blood cells is the gold standard method, but the result depends on skill and experience of the hematologist. In this paper we present a comparative study of feature extraction using two approaches for classifcation of white blood cells. In the frst approach, features were extracted using traditional image processing method and in the second approach we employed AlexNet which is a pre-trained convolutional neural network as feature generator. We used neural network for classifcation of WBCs. The results demonstrate that, clas- sifcation result is slightly better for the features extracted using the convolutional neural network approach compared to traditional image processing approach. The average accuracy and sensitivity of 99% was obtained for classifcation of white blood cells. Hence, any one of these methods can be used for classifcation of WBCs depending availability of data and required resources. Keywords Peripheral blood smear analysis · White blood cells · Classifcation · Deep learning · Decision support system · Computer aided detection Introduction Peripheral blood smear (PBS) analysis is a common labora- tory procedure to assess health condition of a person. White blood cell (WBC) analysis plays a vital role in diagnosing many diseases such as leukemia, lymphoma, neutrophilia, eosinophilia, infections etc. There are fve types of WBCs namely lymphocyte, monocyte, neutrophil, eosinophil and basophil as shown in Fig. 1a–e. They vary in color, shape, size and texture. Abnormality of WBCs can be present in two ways namely count related disorders and morphology based disorders. Complete blood count (CBC) and diferen- tial count (DC) are usually considered in the laboratories to diagnose count related disorders. CBC is performed to pro- vide the total WBC count which includes total count of fve types WBCs whereas DC is performed to provide the count of each type of WBCs. Also morphological analysis such as shape, color and size of WBCs are carried out to diagnose diseases such as leukemia, lymphoma etc. WBCs vary in size, shape and color in abnormal conditions as shown in Fig. 1f–j. Abnormal WBCs in the fgure include myelob- lat, lymphoblat and degenerated WBC. These are immature WBCs and presence of these cells in peripheral blood indi- cate leukemia. Analysis of WBCs is a laborious procedure and it is burden on hematologists. Automation of detection of WBCs would reduce the workload of hematologists. Also, automated classifcation of WBCs will lead to faster diagno- sis and objective results. Automation of WBC classifcation has been addressed since last fve decades using traditional image processing approach. This approach involves image processing pipeline consisting of segmentation, feature extraction and classif- cation. These are interdependent steps, hence segmentation of required region and selected features greatly afect the classifcation results. WBC segmentation is one of the most challenging tasks in medical image processing due to its complex biological appearance, inconsistent staining and * Keerthana Prasad keerthana.prasad@manipal.edu 1 School of Information Sciences, MAHE, Manipal, India 2 Department of ECE, NMAMIT, NITTE, Karkala, India 3 Kasturba Medical College, MAHE, Manipal, India