Original ArticleWhite Blood Cells Image Classification Using Deep Learning with Canonical Correlation Analysis
Graphical abstract
Introduction
Blood cells primarily consist of red blood cells (RBCs), white blood cells (WBCs) and platelets [1]. In blood, WBCs contribute significant role in the human body's immune system, moreover, it is also referred to as an immune cell. WBC defends the body against epidemic diseases and extrinsic invaders. Usually, hematologists use coarse knowledge in medical application to classify white blood cells into two groups (a) granular cells: neutrophil, eosinophil, basophil and (b) non-granular cells: monocyte and lymphocyte [2]. Doctors consistently use this primitive data as standards for deciding the seriousness of blood disease. Thus, the research of white blood cell classification has become an important and essential need in the medical analysis.
Each kind of white blood cell is defined with a particular defending role to resist against foreign species. The Neutrophil is one of the main white blood cells which protects against bacteria. Eosinophils destroy parasites and perform a task in allergies whereas the basophils function in allergic reactions. Monocyte enters the cell to destroy the damaged tissues from the body. Lymphocytes are complicated cells to balance cell-mediated immunity. Lymphocytes are totally different from the other WBC, as a result, they memorize invasive micro-organisms and viruses. Fig. 1, Fig. 2.
Different blood cell classification algorithms are primarily focused on white blood cell classification. According to WBC features, appearances, textures, patterns and various attributes extracted by the feature engineering technique in deep learning plays a principal role in image classification. At present, image processing techniques have introduced for classification, segmentation, feature extraction, and identification [4]. The model previously introduced to classify the WBCs images by hematology analyzer was manual process and time-consuming. Also, it introduces errors in detection. Computer-aided diagnostic machine learning algorithms based on image processing like k-means clustering, decision tree, support vector machine (SVM), and merging of CNN and RNN [3] are compared in Table 1. These models can be enhanced further to classify and to produce higher accuracy rates.
The objective of this research is to improve the classification accuracy employing multiple nuclei patch by overlapping features of cells simultaneously, rather than extracting single cell nuclei patch from WBC subtypes. Deep Learning model learns, trains, validates and regenerates the features from multiple images by itself. This innovative approach of overlapping nuclei performed using the canonical correlation analysis techniques helps to extract the features and patches from input images. CCA method helps to enhance the recognition rate, reduce computational time and overfitting problems [5]. Further, CCA trained images are applied to CNN which converges with RNN for better classification accuracy.
This paper is organized as follows: In section 2, background review and some related works in the field is discussed. Section 3 presents the techniques and methodology followed to accomplish the research goals. In section 4, the experimental results are presented along with the discussion. The article is concluded in last section 5.
Section snippets
Review of related literature
In this section, various algorithms of white blood classification are briefly described. Rosyadi et al. introduced an analysis that has the capability of WBC image classification by considering blood smear samples using the digital microscope [12]. The authors further adopted the Ostu threshold technique and KNN grouping method for classification. Depending on their investigation it was finalized that after implementation of k-means clustering to classify WBC, configured feature generated
Methodology
This section illustrates the approach for the classification of WBC image samples utilizing the canonical correlation analysis method. The flowchart for the proposed algorithm is depicted in Fig. 3. In this work, input image feature extraction is performed with the CCA technique to obtain a higher classification rate.
The most important and powerful addition in the network is the CCA, that has achieved better detection accuracy and introduced by Hotelling [21]. In the CCA network images are
Experimental results and discussion
The canonical correlation analysis approach is applied over the entire dataset of blood cell images before undergoing training and validating process through convolutional neural network and recurrent neural network. Instead of single cell images, it considers multiple cell images due to which higher features are extracted and results in high accuracy using CCA technique.
Conclusion
This article presents canonical correlation analysis based deep learning architecture by combining CNN and LSTM for blood cell image classification task. CCA extracts variety of overlapped features from input image thereby enhancing the accuracy rate as compared to other similar deep learning algorithms. Fine-tuning and transfer learning is used to train the WBCs dataset. The results evaluated from proposed model have been satisfactory, showing better precision, sensitivity, specificity and
Human and animal rights
The authors declare that the work described has been carried out in accordance with the Declaration of Helsinki of the World Medical Association revised in 2013 for experiments involving humans as well as in accordance with the EU Directive 2010/63/EU for animal experiments.
Informed consent and patient details
The authors declare that this report does not contain any personal information that could lead to the identification of the patient(s).
Funding
This work did not receive any grant from funding agencies in the public, commercial, or not-for-profit sectors.
Author contributions
All authors attest that they meet the current International Committee of Medical Journal Editors (ICMJE) criteria for Authorship.
CRediT authorship contribution statement
A.M. Patil: Conceptualization, Data curation, Formal analysis, Methodology, Resources, Software, Validation, Writing - original draft. M.D. Patil: Formal analysis, Methodology, Project administration, Resources, Supervision, Writing - review & editing. G.K. Birajdar: Conceptualization, Data curation, Formal analysis, Methodology, Supervision, Validation, Visualization.
Declaration of Competing Interest
The authors declare that they have no known competing financial or personal relationships that could be viewed as influencing the work reported in this paper.
References (31)
- et al.
Canonical correlation analysis networks for two-view image recognition
Inf Sci
(2017) - et al.
Convolutional neural networks for hyperspectral image classification
Neurocomputing
(2017) - et al.
Recognition of peripheral blood cell images using convolutional neural networks
Comput Methods Programs Biomed
(2019) The classification of white blood cells
IEEE Trans Biomed Eng
(1972)- et al.
Segmentation and classification of white blood cells
(2000) - et al.
Combining convolutional neural network with recursive neural network for blood cell image classification
IEEE Access
(2018) - et al.
Imagenet classification with deep convolutional neural networks
Adv Neural Inf Process Syst
(2012) - et al.
Classification of RBC and WBC in peripheral blood smear using KNN
Paripex - Indian J Res
(2013) - et al.
Support vector machines applied to white blood cell recognition
- et al.
Prediction and diagnosis of leukemia using classification algorithms
Automatic classification of leukocytes using morphological features and naive Bayes classifier
Automatic detection and classification of leukocytes using convolutional neural networks
Med Biol Eng Comput
Recurrent residual convolutional neural network based on U-Net (R2U-Net) for medical image segmentation
Classification of leukocyte images using k-means clustering based on geometry features
Malaria parasite detection in giemsa-stained blood cell images
Cited by (85)
Harnessing deep learning for blood quality assurance through complete blood cell count detection
2024, e-Prime - Advances in Electrical Engineering, Electronics and EnergyDeep integrated fusion of local and global features for cervical cell classification
2024, Computers in Biology and MedicineLMNS-Net: Lightweight Multiscale Novel Semantic-Net deep learning approach used for automatic pancreas image segmentation in CT scan images
2023, Expert Systems with ApplicationsAutomatic base-model selection for white blood cell image classification using meta-learning
2023, Computers in Biology and MedicineWhite blood cell automatic classification using deep learning and optimized quaternion hybrid moments
2023, Biomedical Signal Processing and ControlWhite blood cells classification using multi-fold pre-processing and optimized CNN model
2024, Scientific Reports