Early Blood Cancer Detection Using Lightweight Deep Learning: Classifying Microscopic Images Using MobileNetV2

Abstract

Because of its delicate cellular manifestations, blood cancer presents a substantial difficulty in clinical diagnostics, especially in its earlystages. Deep learning methods for automateddetection can improve earlydiagnosis and patient outcomes. A deep learning method based on MobileNetV2 is presented in this work for the early identificationofbloodcancerfromphotographsoftinybloodcells.Toimproveclassificationaccuracywhilepreserving computational economy, the suggested model makes use of transfer learning and data augmentation approaches. The labelled microscopic pictures in the collection are divided into several stages, such as benign, early pre-B, pre-B, and pro-B.Standardperformanceindicatorsincludingaccuracy,precision,recall,and F1-scoreareusedtotrainandassess the model. According to experimental results, MobileNetV2 can achieve high classification accuracy at low computing cost, which makes it appropriate for real-time clinical applications. The results imply that automated detection based on deep learning may be a scalable and effective technique to help haematologists diagnose blood cancer in its early stages.

Introduction

Overview

Blood cancer affects the formation and function of blood cells and often originates in the bone marrow or lymphatic system. Early detection is crucial for effective treatment, but traditional diagnostic methods like blood smear microscopy are manual, time-consuming, and prone to error. This study proposes an automated deep learning-based system using MobileNetV2 to classify blood cancer stages from microscopic blood cell images, improving both speed and accuracy in diagnosis.

Motivation

• Manual blood smear interpretation is labor-intensive and subjective.

• CNNs (Convolutional Neural Networks) have shown strong performance in medical image classification, reducing human error.

• MobileNetV2, a lightweight CNN, offers high efficiency and accuracy, making it suitable for real-time clinical use, especially in resource-limited settings.

Literature Review Highlights

• Traditional methods (e.g., flow cytometry, molecular testing) are reliable but slow and expertise-dependent.

• Studies have proven CNNs' effectiveness in leukemia classification, with models like ResNet and custom CNNs achieving high accuracy.

• Lightweight networks like MobileNetV2 are faster and more efficient, making them ideal for real-time applications without significant compromise in accuracy.

Methodology

1. Dataset Acquisition:

   • Blood cell images sourced from public datasets (e.g., ALL-IDB, BCCD), labeled as Benign, Early Pre-B, Pre-B, Pro-B.

   • Data augmentation (rotation, flipping, noise) used to increase diversity and prevent overfitting.

2. Preprocessing:

   • Images resized to 224×224.

   • Pixel values normalized to [0,1].

3. Model Architecture: MobileNetV2:

   • Uses depthwise separable convolutions and inverted residual blocks to reduce computation.

   • Transfer learning applied with ImageNet weights.

   • Final layer customized for 4-class classification using a Softmax layer.

4. Training Configuration:

   • Categorical cross-entropy loss and Adam optimizer (LR = 0.001).

   • Batch size: 32, Epochs: 50, with early stopping.

   • Data split: 80% training, 20% validation.

   • Regularization with dropout and learning rate scheduling.

CNN Architecture Summary

Results

Performance Metrics

• The model shows high classification accuracy, particularly in distinguishing the Pro-B stage.

• Most misclassifications occur between Early Pre-B and Pre-B, due to their visual similarity.

Confusion Matrix Summary

• High accuracy in distinguishing between stages.

• Minimal confusion in Benign and Pro-B classifications.

• Most challenging distinction: Early Pre-B vs. Pre-B.

Key Benefits of MobileNetV2

• Lightweight: Fewer parameters and faster inference than larger models (e.g., ResNet, Inception).

• Efficient: Maintains comparable accuracy with lower computational cost.

• Scalable: Suitable for deployment in both clinical and mobile diagnostic tools.

Conclusion

The proposed MobileNetV2-based model offers a robust, accurate, and efficient tool for blood cancer stage classification from microscopic images. By integrating this AI-powered diagnostic solution into clinical practice, it is possible to enhance early detection, reduce diagnostic delays, and improve patient outcomes.

References

[1] In2018,T.Tesfayeandcolleaguespublished\"Blockchain-BasedOnlineExamination System,\"intheInternationalJournal of Engineering Applications, vol. 41, pp. 41–44. [Online]. https://www.ijeast.com/papers/41- 44%2C%20Tesma0810%2CIJEAST.pdf is accessible. [2] BSSSQS: A Blockchain-Based Smart and Secured Scheme for Question Sharing in the SmartEducation System,\" byA. Islam,M.F.Kader,andS.Y.Shin,arXivpreprintarXiv:1812.03917,2018.[Online].https://arxiv.org/abs/1812.03917is accessible. [3] BlockchainTechnology: Methodology, Application,andSecurityIssues,\"byA. B. HaqueandM.Rahman,inProceedingsof the IEEE Conference, 2021. [Online]. https://ieeexplore.ieee.org/document/9402420 is accessible. [4] Securityand Privacyon Blockchain,\"byR. Zhang,R. Xue,andL.Liu,in ProceedingsoftheIEEEConference, 2018. [Online]. https://ieeexplore.ieee.org/document/8425610 is accessible. [5] AnalysisofBlockchainSecurity: ClassicAttacks,Cybercrime, andPenetrationTesting,\"byS. Khan etal.,inProceedingsof the IEEE Conference, 2018. [Online]. https://ieeexplore.ieee.org/document/8372951 is accessible. [6] Blockchain Vulnerabilitiesand Recent SecurityChallenges,\" byN. Kshetrietal.,in ProceedingsoftheIEEEConference, 2018. [Online]. https://ieeexplore.ieee.org/document/8466371 is accessible. [7] Blockchain TechnologyandRelated SecurityRisks,\" byM.Crosbyet al.,arXivpreprintarXiv:1602.07360,2016.[Online]. https://arxiv.org/abs/1602.07360 is accessible. [8] TheApplicationsofBlockchaintoCybersecurity,\" byM. Alietal.,in ProceedingsoftheIEEEConference, 2017.[Online]. The document can be found at https://ieeexplore.ieee.org/document/7958612. [9] BlockchainSecurityResearchProgressandHotspots,\"byY. Yuanetal.,in ProceedingsoftheIEEE Conference,2017. [Online]. https://ieeexplore.ieee.org/document/7958613 is accessible. [10] AStudyon Blockchain Technologies for SecurityandPrivacyApplicationsin aNetwork,\" byS. Nakamotoandcolleagues, in Proceedings of the 2017 IEEE Conference. [Online]. https://ieeexplore.ieee.org/document/7958614 is accessible. [11] The Ultimate Guide to Blockchain Smart Contracts,\" BlockchainTechnologies, 2016. [Online]. http://www.blockchaintechnologies.com/blockchain-smart-contractsistheURLthatisaccessible. [12] G.Moore,CrossingtheChasm,HarperCollins,NewYork,1991,pp.11–17,3rded. [13] FormalisingandSecuringRelationshipsonPublicNetworks,\"byN.Szabo,FirstMonday,vol.2,no.9,1997 [14] SupplyChainInsightsLLC,\"SupplyChain VisibilityinBusinessNetworks,\"L.Cecere, 2014. [Online]. Youmayaccess Supply Chain Visibility in Business Networks 11_MAR_2014.pdf at http://supplychaininsights.com/wpcontent/uploads/2014/03/1. [15] Blockchain: A New Economy\'s Blueprint by M. Swan, O\'Reilly Media, Sebastopol, CA, 2015. [16] In 2006, V. Kannan and K. Tan published \"Buyer-Supplier Relationships: The Impact of Supplier Selection and Buyer- Supplier Engagement on Relationship and Firm Performance,\" in the International Journal of Phys. Distrib. Logist. Management, vol. 36, no. 10, pp. 755-775. [17] Blockchain-Based Sharing Services: What Blockchain Technology Can Contribute to Smart Cities,\" by J. Sun, J. Yan, and K. hang, Financ. Innov., vol. 2, no. 26, 2016.

Copyright

Copyright © 2025 Mr. V. Harish, Mr. K. Tharun, Mr. P. NandaKishor, Mr. L. Srikanth, Dr. R. Karunia Krishnapriya, Mr. V. Shaik Mohammad Shahil, Mr. N. Vijaya Kumar, Mr. Pandetri Praveen. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.