Improved White Blood Cells Classification based on Pre-trained Deep Learning Models

Published online: Mar 18, 2020 Full Text: PDF (2.02 MiB) DOI: 10.24138/jcomss.v16i1.818
Cite this paper
Ensaf H. Mohamed, Wessam H. El-Behaidy, Ghada Khoriba, Jie Li


Leukocytes, or white blood cells (WBCs), are microscopic organisms that fight against infectious disease, bacteria, viruses, and others. The manual method to classify and count WBCs is tedious, time-consuming and may has inaccurate results, whereas the automated methods are costly. The objective of this work is to automatically identify and classify WBCs in a microscopic image into four types with higher accuracy. BCCD is the used dataset in this study, which is a scaleddown blood cell detection dataset. BCCD is firstly pre-processed by passing through several processes such as segmentation and augmentation,then it is passed to the proposed model. Our model combines the privilege of deep models in automatically extracting features with the higher classification accuracy of traditional machine learning classifiers.The proposed model consists of two main layers; a shallow tuning pre-trained model and a traditional machine learning classifier on top of it. Here, ten different pretrained models with six different machine learning are used in this study. Moreover, the fully connected network (FCN) of pretrained models is used as a baseline classifier for comparison. The evaluation process shows that the hybrid between MobileNet-224 as feature extractor with logistic regression as classifier has a higher rank-1 accuracy with 97.03%. Besides, the proposed hybrid model outperformed the baseline FCN with 25.78% on average.


Deep learning, feature extraction, classification, White Blood Cells (WBCs)
Creative Commons License 4.0
This work is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License.