Skip to main content
SpringerLink
Log in

Leukemia classification using different CNN-based algorithms-comparative study

  • Original Article
  • Published:
Neural Computing and Applications Aims and scope Submit manuscript

Abstract

Leukemia or blood cancer has its roots in the bone marrow. It is distinguished by irregular white blood cell proliferation. Early diagnosis of leukemia is crucial to increase the effectiveness of its treatment. However, manual methods to detect and classify leukemia from blood microscopic images are time-consuming and susceptible to inter and intra-observer variations. Therefore, a low-cost, fully automated, and robust system for leukemia detection and classification is required. Many algorithms have been found in the literature to detect it but not to classify its four different types with high accuracy. The proposed study uses different CNN-based algorithms to detect leukemia and classify its types. AlexNet, DenseNet, ResNet, and VGG16 were used. Images from three datasets were tested; 108 images from the ALL-IDB dataset, 547 images ASH Image bank, and 15 images captured in the biomems and bionanotechnology laboratory at JUST. The best results were achieved by retraining a pre-trained model through transfer learning with fine-tuning weights. All models used gave acceptable accuracies, reaching 99.8%, 99.7%, and 94% for training, validation, and testing sets, respectively. The proposed study provides clear, accurate, and reliable guidance to researchers who are working on leukemia detection and classification, and hence provides the medical staff with an easy and effective system to diagnose leukemia without any human intrusion; furthermore, it is expected to save time and effort at a lower price.

This is a preview of subscription content, log in via an institution to check access.

Access this article

Log in via an institution

Price excludes VAT (USA)
Tax calculation will be finalised during checkout.

Instant access to the full article PDF.

Institutional subscriptions

Fig. 1
Fig. 2
Fig. 3
Fig. 4
Fig. 5

Similar content being viewed by others

Data availability

The authors declare that the data supporting the findings of this study are available in its repository: the ALL-IDB dataset at: https://scotti.di.unimi.it/all/#datasets, ASH Image bank at https://libraries.usc.edu/databases/american-society-hematology-ash-image-bank, and images captured in biomems and bionanotechnology laboratory at JUST are available from the corresponding author upon reasonable request.

Notes

  1. More details about the proposed models are in Appendix A.

References

  1. Siegel RL, Miller KD, Fuchs HE, Jemal A (2022) Cancer statistics. CA A Cancer J Clin 72(1):7–33

    Article  Google Scholar 

  2. ‘Leukemia: Symptoms, Types, Causes & Treatments’. Accessed: Nov. 08, 2021. [Online]. Available: https://my.clevelandclinic.org/health/diseases/4365-leukemia

  3. ‘French-American-British Classification’. Accessed: Nov. 08, 2021. [Online]. Available: https://datadictionary.nhs.uk/nhs_business_definitions/french-american-british_classification.html

  4. Catovsky D, Sultan C, Bennett JM (1977) Classification of acute leukemia. Ann Intern Med 87(6):740–753

    Article  Google Scholar 

  5. Mensen VT et al (2017) Development of cortical thickness and surface area in autism spectrum disorder. NeuroImage Clin 13:215–222. https://doi.org/10.1016/j.nicl.2016.12.003

    Article  Google Scholar 

  6. Haralick RM, Shanmugam K, Dinstein IH (1973) Textural features for image classification. IEEE Trans Syst Man Cybern 6:610–621

    Article  Google Scholar 

  7. Pentland AP (1984) Fractal-based description of natural scenes. IEEE Trans Pattern Anal Mach Intell 6:661–674

    Article  Google Scholar 

  8. Pentland A (1983) Fractal-Based Description. In: IJCAI (pp. 973-981).

  9. Pearson K (1901) LIII On lines and planes of closest fit to systems of points in space. Lond Edinb Dublin Philos Mag J Sci 2(11):559–572

    Article  Google Scholar 

  10. Wold S, Esbensen K, Geladi P (1987) Principal component analysis. Chemom Intell Lab Syst 2(1–3):37–52

    Article  Google Scholar 

  11. Lourakis MIA (2005) A brief description of the Levenberg-Marquardt algorithm implemented by levmar. Found Res Technol 4(1):1–6

    Google Scholar 

  12. RD Labati, V Piuri, and F Scotti, (2011) All-IDB: The acute lymphoblastic leukemia image database for image processing. In: 2011 18th IEEE International Conference on Image Processing, 2045–2048.

  13. Madhukar M, Agaian S, Chronopoulos AT (2012). New decision support tool for acute lymphoblastic leukemia classification. In: Image processing: Algorithms and systems X; and parallel processing for imaging applications II (Vol. 8295, pp. 367-378). SPIE.

  14. Cortes C, Vapnik V (1995) Support-vector networks. Mach Learn 20(3):273–297

    Article  Google Scholar 

  15. Agaian S, Madhukar M, Chronopoulos AT (2018) A new acute leukaemia-automated classification system. Comput Methods Biomech Biomed Eng Imaging Vis 6(3):303–314

    Article  Google Scholar 

  16. Khobragade S, Mor DD, Patil CY (2015). Detection of leukemia in microscopic white blood cell images. In: 2015 international conference on information processing (ICIP) (pp. 435-440). IEEE.

  17. M Castelluccio, G Poggi, C Sansone, L Verdoliva, (2015) Land use classification in remote sensing images by convolutional neural networks’, ArXiv Prepr. ArXiv150800092,

  18. D Wang, A Khosla, R Gargeya, H Irshad, AH Beck, (2016) Deep learning for identifying metastatic breast cancer’, ArXiv Prepr. ArXiv160605718,

  19. Kumar A, Kim J, Lyndon D, Fulham M, Feng D (2016) An ensemble of fine-tuned convolutional neural networks for medical image classification. IEEE J Biomed Health Inform 21(1):31–40

    Article  Google Scholar 

  20. Wang Z, Wang W, Yang Y, Han Z, Xu D, Su C (2022) CNN- and GAN-based classification of malicious code families: a code visualization approach. Int J Intell Syst 37(12):12472–12489. https://doi.org/10.1002/int.23094

    Article  Google Scholar 

  21. Thanh TTP, Vununu C, Atoev S, Lee S-H, Kwon K-R (2018) Leukemia blood cell image classification using convolutional neural network. Int J Comput Theory Eng 10(2):54–58

    Article  Google Scholar 

  22. Loey M, Naman M, Zayed H (2020) Deep transfer learning in diagnosing Leukemia in blood cells. Computers 9(2):29

    Article  Google Scholar 

  23. A Krizhevsky, I Sutskever, GE Hinton, (2012) Imagenet classification with deep convolutional neural networks. In: Advances in neural information processing systems, pp. 1097–1105.

  24. ‘ImageBank | Home | Regular Bank’. Accessed: Feb. 07, 2022. [Online]. Available: https://imagebank.hematology.org/

  25. ‘Blood Cell Images | Kaggle’. Accessed: Feb. 12, 2022. [Online]. Available: https://www.kaggle.com/paultimothymooney/blood-cells

  26. Vogado LH, Veras RDM, Andrade AR, de Araujo FH, Silva RR, Aires KR (2017). Diagnosing leukemia in blood smear images using an ensemble of classifiers and pre-trained convolutional neural networks. In: 2017 30th SIBGRAPI Conference on Graphics, Patterns and Images (SIBGRAPI) (pp. 367-373). IEEE.

  27. Y. Jia et al., (2014) Caffe: Convolutional architecture for fast feature embedding, In: Proceedings of the 22nd ACM international conference on Multimedia, 675–678.

  28. K Chatfield, K Simonyan, A Vedaldi, A Zisserman, (2014) Return of the devil in the details: Delving deep into convolutional nets’, ArXiv Prepr. ArXiv14053531.

  29. Popescu M-C, Balas VE, Perescu-Popescu L, Mastorakis N (2009) Multilayer perceptron and neural networks. WSEAS Trans Circuits Syst 8(7):579–588

    Google Scholar 

  30. Cutler A, Cutler DR, Stevens JR (2012) Random forests. Methods and applications, Ensemble machine learning, pp 157–175

    Google Scholar 

  31. Breiman L (2001) Random forests. Mach Learn 45:5–32

    Article  Google Scholar 

  32. Vogado LHS, Veras RMS, Araujo FHD, Silva RRV, Aires RT (2018) Engineering applications of artificial intelligence leukemia diagnosis in blood slides using transfer learning in CNNs and SVM for classification. Eng Appl Artif Intell 72(April):415–422. https://doi.org/10.1016/j.engappai.2018.04.024

    Article  Google Scholar 

  33. ‘Transfer Learning for Deep Learning | Engineering Education (EngEd) Program | Section’. Accessed: Nov. 08, 2021. [Online]. Available: https://www.section.io/engineering-education/transfer-learning-with-deep-learning/

  34. ‘A Gentle Introduction to Transfer Learning for Deep Learning’. Accessed: Nov. 08, 2021. [Online]. Available: https://machinelearningmastery.com/transfer-learning-for-deep-learning/

  35. Rollins-Raval M et al (2013) CD123 immunohistochemical expression in acute myeloid leukemia is associated with underlying FLT3-ITD and NPM1 mutations. Appl Immunohistochem Mol Morphol 21(3):212–217

    Article  Google Scholar 

  36. Sarrafzadeh O, Dehnavi A (2015) Nucleus and cytoplasm segmentation in microscopic images using K-means clustering and region growing. Adv Biomed Res 4(1):174–174

    Article  Google Scholar 

  37. Sarrafzadeh O, Rabbani H, Talebi A, Banaem HU (2014). Selection of the best features for leukocytes classification in blood smear microscopic images. In: Medical Imaging 2014: Digital Pathology (Vol. 9041, pp. 159-166). SPIE.

  38. Shafique S, Tehsin S (2018) Acute lymphoblastic leukemia detection and classification of its subtypes using pretrained deep convolutional neural networks. Technol Cancer Res Treat 17:1533033818802789. https://doi.org/10.1177/1533033818802789

    Article  Google Scholar 

  39. Rehman A, Abbas N, Saba T, Rahman SIU, Mehmood Z, Kolivand H (2018) Classification of acute lymphoblastic leukemia using deep learning. Micro Res Tech 81(11):1310–1317

    Article  Google Scholar 

  40. Ahmed N, Yigit A, Isik Z, Alpkocak A (2019) Identification of leukemia subtypes from microscopic images using convolutional neural network. Diagnostics 9(3):104

    Article  Google Scholar 

  41. ‘American Society of Hematology - Hematology.org’. Accessed: Nov. 08, 2021. [Online]. Available: https://www.hematology.org/

  42. S Pandya, TR Gadekallu, PK Reddy, WWang, M Alazab, (2022) InfusedHeart: a novel knowledge-infused learning framework for diagnosis of cardiovascular events, In: IEEE Trans Comput Soc Syst, 1–10, https://doi.org/10.1109/TCSS.2022.3151643.

  43. Horobin RW (2011) How Romanowsky stains work and why they remain valuable—including a proposed universal Romanowsky staining mechanism and a rational troubleshooting scheme. Biotech Histochem 86(1):36–51

    Article  Google Scholar 

  44. ‘Cross-Validation in Machine Learning: How to Do It Right - neptune.ai’. Accessed: Mar. 09, 2022. [Online]. Available: https://neptune.ai/blog/cross-validation-in-machine-learning-how-to-do-it-right

  45. ‘Cross Validation’. Accessed: Mar. 09, 2022. [Online]. Available: https://www.cs.cmu.edu/~schneide/tut5/node42.html

  46. ‘Transfer learning and fine-tuning | TensorFlow Core’. Accessed: Dec. 05, 2021. [Online]. Available: https://www.tensorflow.org/tutorials/images/transfer_learning

  47. ‘What Is Transfer Learning? A Simple Guide | Built In’. Accessed: Dec. 05, 2021. [Online]. Available: https://builtin.com/data-science/transfer-learning

  48. ‘ImageNet’. Accessed: Feb. 07, 2022. [Online]. Available: https://www.image-net.org/

  49. ‘Transfer learning from pre-trained models | by Pedro Marcelino | Towards Data Science’. Accessed: Dec. 05, 2021. [Online]. Available: https://towardsdatascience.com/transfer-learning-from-pre-trained-models-f2393f124751

  50. ‘AlexNet: The Architecture that Challenged CNNs | by Jerry Wei | Towards Data Science’. Accessed: Dec. 05, 2021. [Online]. Available: https://towardsdatascience.com/alexnet-the-architecture-that-challenged-cnns-e406d5297951

  51. ‘ImageNet’. Accessed: Feb. 07, 2022. [Online]. Available: https://image-net.org/challenges/LSVRC/2014/

  52. ‘VGG16 - Convolutional Network for Classification and Detection’. Accessed: Dec. 05, 2021. [Online]. Available: https://neurohive.io/en/popular-networks/vgg16/

  53. ‘7.6. Residual Networks (ResNet) — Dive into Deep Learning 0.17.0 documentation’. Accessed: Dec. 05, 2021. [Online]. Available: https://d2l.ai/chapter_convolutional-modern/resnet.html

  54. ‘Architecture of DenseNet-121’. Accessed: Dec. 05, 2021. [Online]. Available: https://iq.opengenus.org/architecture-of-densenet121/

  55. Jabbar H, Khan RZ (2015) Methods to avoid over-fitting and under-fitting in supervised machine learning (comparative study). Comput Sci, Commun Instrument Dev 70(10.3850):978–981. https://doi.org/10.3850/978-981-09-5247-1_017

    Article  Google Scholar 

  56. Yadav S, Shukla S (2016). Analysis of k-fold cross-validation over hold-out validation on colossal datasets for quality classification. In: 2016 IEEE 6th International conference on advanced computing (IACC) (pp. 78-83). IEEE.https://doi.org/10.1109/IACC.2016.25.

  57. Pal K, Patel BV (2020). Data classification with k-fold cross validation and holdout accuracy estimation methods with 5 different machine learning techniques. In: 2020 fourth international conference on computing methodologies and communication (ICCMC) (pp. 83-87). IEEE.https://doi.org/10.1109/ICCMC48092.2020.ICCMC-00016.

  58. Poojary R, Raina R, Mondal AK (2021) Effect of data-augmentation on fine-tuned CNN model performance. IAES Int J Artif Int 10(1):84

    Google Scholar 

  59. Anwar T, Zakir S (2021) ‘Effect of image augmentation on ECG Image classification using deep learning. Int Conf Artif Intell ICAI 2021:182–186

    Google Scholar 

  60. Arivuselvam B, Sudha S (2022) Leukemia classification using the deep learning method of CNN. J X-Ray Sci Technol 30(3):567–585. https://doi.org/10.3233/XST-211055

    Article  Google Scholar 

  61. ‘ImageNet’. https://image-net.org/challenges/LSVRC/2012/ (accessed Feb. 07, 2022).

  62. ‘AlexNet - ImageNet Classification with Convolutional Neural Networks’. https://neurohive.io/en/popular-networks/alexnet-imagenet-classification-with-deep-convolutional-neural-networks/ (accessed Dec. 05, 2021).

  63. ‘AlexNet: The Architecture that Challenged CNNs | by Jerry Wei | Towards Data Science’. https://towardsdatascience.com/alexnet-the-architecture-that-challenged-cnns-e406d5297951 (accessed Dec. 05, 2021).

  64. ‘ImageNet’. https://image-net.org/challenges/LSVRC/2014/ (accessed Feb. 07, 2022).

  65. K Simonyan and A Zisserman, (2014) Very deep convolutional networks for large-scale image recognition’, ArXiv Prepr. ArXiv14091556

  66. ‘VGG16 - Convolutional Network for Classification and Detection’. https://neurohive.io/en/popular-networks/vgg16/ (accessed Dec. 05, 2021).

  67. He K, Zhang X, Ren S, Sun J (2016). Deep residual learning for image recognition. In: Proceedings of the IEEE conference on computer vision and pattern recognition (pp. 770-778).

  68. ‘7.6. Residual Networks (ResNet) — Dive into Deep Learning 0.17.0 documentation’. https://d2l.ai/chapter_convolutional-modern/resnet.html (accessed Dec. 05, 2021).

  69. Huang G, Liu Z, Van Der Maaten L, Weinberger KQ (2017). Densely connected convolutional networks. In: Proceedings of the IEEE conference on computer vision and pattern recognition (pp. 4700-4708).https://doi.org/10.1109/CVPR.2017.243.

  70. ‘Architecture of DenseNet-121’. https://iq.opengenus.org/architecture-of-densenet121/ (accessed Dec. 05, 2021).

Download references

Funding

The project was supported by the Deanship of Research at Jordan University of Science and Technology, Irbid, Jordan. Project # 20210058 and 20180369.

Author information

Authors and Affiliations

  1. Biomedical Engineering Department, Jordan University of Science and Technology, P.O.Box 3030, Irbid, 22110, Jordan

    Areen K. Al-Bashir, Ruba E. Khnouf & Lamis R. Bany Issa

Authors
  1. Areen K. Al-Bashir
    View author publications

    You can also search for this author in PubMed Google Scholar

  2. Ruba E. Khnouf
    View author publications

    You can also search for this author in PubMed Google Scholar

  3. Lamis R. Bany Issa
    View author publications

    You can also search for this author in PubMed Google Scholar

Contributions

All authors contributed to the study's conception and design. Material preparation, data collection, and analysis were performed by AKAB, REK, and LRR. BI. The first draft of the manuscript was written by AKLB and all authors commented on previous versions of the manuscript. All authors read and approved the final manuscript.

Corresponding author

Correspondence to Areen K. Al-Bashir.

Ethics declarations

Conflict of interest

The authors have no relevant financial or non-financial interests to disclose.

Ethics approval

NA.

Consent to participate

NA.

Consent to publish

NA.

Additional information

Publisher's Note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Appendix A

Appendix A


Description of the pre-trained models in the study: AlexNet, VGG16, ResNet, and DenseNet.


A. Alexnet


AlexNet is a convolutional neural network (CNN) that has had a significant impact on machine learning, particularly in the application of deep learning to machine vision. AlexNet won the ImageNet Large Scale Visual Recognition Challenge in 2012 [61].

Use: AlexNet demonstrated how a deep convolutional neural network can be used to solve image classification problems.

Architecture: AlexNet has eight weighted layers, the first five of which are convolutional and the last three of which are fully connected. The last fully connected layer's output is sent into a 1000-way softmax, which generates a distribution across the 1000 class labels. After each convolutional and fully connected layer, Rectified Linear Units (ReLU) is applied. A dropout layer is added before the first and second fully connected years. In a forward pass, the network comprises 62.3 million parameters and requires 1.1 billion computing units. Convolution layers, which account for 6% of all parameters, take about 95% of the process. Figure 6 shows Alexnet model architecture. 

Fig. 6
figure 6

Alexnet Architecture that is used in the process

Full size image

Unique features: AlexNet is characterized by ReLU nonlinearity that enables CNN to reach a 25% error six times faster than CNN with another function. It also addresses the overlapping pooling to the output of neighboring groups of neurons so the model with overlapping pooling leads to a reduction in error by about 0.5% and it is harder to overfit [62, 63].


B. VGG


VGG16 is a convolutional neural network architecture that won the 2014 ILSVR(ImageNet) competition [64]. It was proposed by K. Simonyan and A. Zisserman in their publication [65]. Using ImageNet, a dataset of over 14 million images belonging to 1000 classes, the model accomplishes 92.7 percent top-5 test accuracy.

Use: It is regarded as one of the best vision model architectures ever created to date. Used for image recognition.

Architecture: The most distinctive feature of VGG16 is that, rather than having a huge number of hyper-parameters, they emphasized having 3 × 3 filter convolution layers with a stride 1, and the same padding is also used in addition to the max pool layer of 2 × 2 filter stride 2. Throughout the architecture, the convolution and max pool layers are arranged in the same way. It has two FC (fully connected layers) in the end, followed by a softmax for output. The 16 in VGG16 relates to the fact that it contains 16 layers with weights. This network is quite huge, with approximately 138 million (estimated) parameters. It outperforms AlexNet by sequentially replacing big kernel-size filters 11 in the first layer and 5 in the second convolutional layers) with numerous 33 kernel-size filters in a sequential fashion [66]. Vgg architecture is depicted in Fig. 7

Fig. 7
figure 7

Vgg16 Architecture that is used in the process

Full size image

Unique features: VGG can reduce the # of parameters in the CONV layers and improve training time.


C. Resnet


ResNet, standing for Residual Network, is a well-known deep learning model that was introduced in a paper titled "Deep Residual Learning for Image Recognition" published in 2015 by Shaoqing Ren, Kaiming He, Jian Sun, and Xiangyu Zhang [67].

Use: ResNet is one of the most widely used and successful deep learning models to date, especially in computer vision applications.

Architecture: ResNet is inspired by VGG’s network followed by a shortcut connection. Two 3 × 3 convolutional layers with the same amount of output channels make up the residual block. A batch normalizing layer and a ReLU activation function follow each convolutional layer. The input is then added right before the final ReLU activation function, skipping these two convolution steps. The output of the two convolutional layers must be of the same size as the input for them to be merged in this configuration [68]. Resnet50, for instance, is a Resnet version of 48 Convolution layers with 1 MaxPool layer and 1 Average Pool layer as shown in Fig. 8

Fig. 8
figure 8

Resnet Architecture that is used in the process

Full size image

Unique features: use of shortcut connections to solve the vanishing gradient problem. On the other hand, shortcuts can improve training time.


D. Densenet


A DenseNet is a convolutional neural network that uses Dense Blocks to connect all layers (with matching feature-map sizes) directly to each other, resulting in dense connections between layers.

Use: Each layer takes further inputs from all previous layers and carries on its feature maps to all following layers to maintain the feed-forward nature [69].

Architecture: DenseNet-121, in summary, has 120 Convolutions, 4 AvgPool, and 1 Fully Connected Layer. All layers, including those within the same dense block and transition layers, distribute their weights over various inputs, allowing deeper layers to make use of characteristics retrieved earlier in the process. Thus, DenseNet-121 has layers 1 Convolution (7 × 7), 58 Convolution (3 × 3), 61 Convolution (1 × 1), 4 AvgPool, and 1 Fully Connected Layer [70]. The Densenet architecture is shown in Fig. 

Fig. 9
figure 9

Densenet Architecture that is used in the process

Full size image

9.

Unique features: the information passed through many layers will not be washed out or vanish by the time it reaches the end of the network.

Rights and permissions

Springer Nature or its licensor (e.g. a society or other partner) holds exclusive rights to this article under a publishing agreement with the author(s) or other rightsholder(s); author self-archiving of the accepted manuscript version of this article is solely governed by the terms of such publishing agreement and applicable law.

Reprints and permissions

About this article

Check for updates. Verify currency and authenticity via CrossMark

Cite this article

Al-Bashir, A.K., Khnouf, R.E. & Bany Issa, L.R. Leukemia classification using different CNN-based algorithms-comparative study. Neural Comput & Applic 36, 9313–9328 (2024). https://doi.org/10.1007/s00521-024-09554-9

Download citation

  • Received:

  • Accepted:

  • Published:

  • Issue Date:

  • DOI: https://doi.org/10.1007/s00521-024-09554-9

Keywords

Search

Navigation