Theuncheckedgrowthofaberrantwhitebloodcells is the hallmark of leukemia, a serious hematological cancer that starts in the bone marrow and bloodstream and causes bleeding, anemia, and weakened immunity. Early and a precise leukemia diagnosisisnecessarytoenhancepatientsurvivalratesand start treatment procedures on time. In this work, a lightweight MobileNetV2 convolutional neural network (CNN) architectureis used to analyze microscopic blood smear images and presentan automated leukemia classification system. To enhance model performance, advanced preprocessing techniques such as LAB color space segmentation, KMeans clustering, and targeted data augmentation were employed to normalize imaging variabilities and highlight leukemic features.
The system was trained and evaluated on a curated dataset comprising 3,242 images from 89 patients, encompassing both benign and malignant cases. Compared to conventional manual examination,thesystemachievessuperiorclassificationaccuracy, precision,andrecall,providingascalableandefficientdiagnostic tool.Notably,MobileNetV2’slightweightarchitectureguarantees quick inference with no processing overhead, which makes it ideal for real-time clinical application, especially in settings with limited resources.
By significantly reducing diagnostic time and minimizing human error, it demonstrates the transformative potential of deep learning and computer vision in hematological diagnostics. Future work will focus on expanding multi-class classification capabilities for different leukemia subtypes and integrating explainableAItechniquestoenhanceclinicalinterpretabilityand trustworthiness.
Introduction
Leukemia, a blood cancer caused by abnormal white blood cell growth, impairs the body’s ability to fight infections and regulate blood functions. Early and accurate diagnosis is crucial but traditional microscopic examination is slow, subjective, and often unavailable in low-resource areas. Advances in artificial intelligence, especially deep learning using Convolutional Neural Networks (CNNs), offer automated, precise analysis of blood smear images.
This research develops a lightweight, efficient mobile-based leukemia detection system using the MobileNetV2 CNN architecture. It enhances image preprocessing through LAB color space conversion and KMeans clustering segmentation to improve feature visibility. The system is implemented as a cross-platform mobile app using Flutter, enabling real-time, portable leukemia screening ideal for underserved regions.
The model was trained on a dataset of 3,242 blood smear images from 89 patients, applying extensive data augmentation and transfer learning to optimize accuracy and robustness. MobileNetV2 achieved outstanding performance metrics (accuracy ~99.6%, precision, recall, and F1-score near 99%), outperforming other lightweight CNNs like EfficientNetB0 and NASNetMobile, while maintaining fast inference times suitable for mobile devices.
The system’s real-time capability, compact size (~10.6 MB), and offline functionality make it particularly valuable for remote healthcare. While minor limitations exist due to variability in staining and imaging, this solution offers a scalable, accessible tool for early leukemia detection, helping reduce diagnostic delays and improve patient outcomes in resource-constrained settings.
Conclusion
This study presents an enhanced approach for automated leukemia detection utilizing the MobileNetV2 architecture, demonstratingsignificantimprovementsinclassificationaccu- racy, recall, and computational efficiency compared to previous methods.Ourmodeloutperformstraditionalmachinelearning- basedsystems,suchasthosedevelopedbyPatilandHiremath
[1]andDasariRajuetal.[2],particularlyindifferentiatingfine morphological features between benign and malignant blood cells.
By integrating an optimized preprocessing pipeline that includesLABcolorspaceconversion,KMeans-basedsegmen- tation, and comprehensive data augmentation strategies, the proposed system enhances feature extraction, leading to more precise classification outcomes. Additionally, the lightweight design of MobileNetV2 ensures its seamless deployment on mobileandedgedevices,makingreal-timeleukemiadetection accessible even in resource-constrained healthcare settings.
The experimental findings confirm the robustness of the proposed system, with the model achieving an accuracy of 99.62%,aprecisionof0.9935,arecallof0.996,andan F1-scoreof0.9945onthevalidationdataset.Furthermore, it maintains a fast inference time of approximately 15.2 milliseconds per image, reinforcing its suitability for real-time clinical applications. Compared to heavier architectures like VGG16 [11] and ensemble learning methods [14], the MobileNetV2 model provides an optimal balance between computational efficiency and diagnostic performance.
Nonetheless, persistent challenges such as variability in blood smear image quality, staining inconsistencies, and limited dataset diversity—issues similarly noted in previous leukemia detection studies [8], [13]—must still be addressed. Futureresearchwillfocusonexpandingthedatasettoincorpo- rate a broader range of imaging conditions, enhancing model robustness, and exploring hybrid deep learning models to furtherimprovediagnosticprecision.Additionally,integrating advanced explainability techniques, such as attention maps, could offer better transparency in model decision-making, aiding clinical adoption.
By addressing these aspects, the proposed system aims to contribute to the development of an efficient, real-time, and scalableAI-drivensolutionforearly-stageleukemiadetection, ultimately supporting timely diagnosis and improving patient outcomes.
References
[1] A. P. Patil and M. Hiremath, ”Machine Learning Model to DetectChronic Leukemia in Microscopic Blood Smear Images,” in *Pro-ceedings of the IEEE International Conference on Smart Systems andInventive Technology (ICSSIT)*, 2022, pp. 450–455.
[2] S. Dasari Raju, K. Krishna Reddy, and D. Kishore Babu, ”Detection ofAcuteMyeloidLeukemiaUsingRandomForestAlgorithm,”in*JournalofInformationTechnologyCaseandApplicationResearch*,vol.15,no.2, pp. 134–140, 2022.
[3] Y. Jusman, R. Mahmod, N. Ghazali, and A. H. Abdullah, ”ComparativeStudy of CNN Architectures for Leukemia Cell Classification: AlexNetversus GoogLeNet,” in *International Journal of Advanced ComputerScienceandApplications(IJACSA)*,vol.11,no.5,pp.456–462,2020.
[4] A. Mishra and N. Patel, ”Image Processing Techniques for LeukemiaCell Detection,” in *International Journal of Innovative Research inComputer and Communication Engineering (IJIRCCE)*, vol. 7, no. 4,
[5] pp.2345–2350,2019.
[6] A. G. Howard, M. Zhu, B. Chen, D. Kalenichenko, et al., ”MobileNets:Efficient Convolutional Neural Networks for Mobile Vision Applica-tions,” arXiv preprint arXiv:1704.04861, 2017.
[7] M. Tan and Q. Le, ”EfficientNet: Rethinking Model Scaling for Con-volutional Neural Networks,” in *Proceedings of the 36th InternationalConference on Machine Learning (ICML)*, vol. 97, pp. 6105–6114,2019.
[8] Z. Zoph, V. Vasudevan, J. Shlens, and Q. Le, ”Learning TransferableArchitecturesforScalableImageRecognition,”in*Proceedingsofthe IEEE Conference on Computer Vision and Pattern Recognition(CVPR)*, 2018, pp. 8697–8710.
[9] M. Ghaderzadeh, F. Asadi, and S. Jafari, ”A Systematic Review ofMachine Learning Techniques for Leukemia Diagnosis Based on BloodCell Images,” *Artificial Intelligence Review*, vol. 55, no. 1, pp.505–543, 2021.
[10] J. R. Devi, P. S. Kadiyala, and L. Kosuri, ”Enhancing Acute Lym-phoblasticLeukemia Classification withCNNModels,” in*IEEEInter-nationalConferenceonDigitalCircuits,CommunicationandElectronics(ICDCCE)*, 2024, pp. 78–83.
[11] N. Jiwani, A. Sharma, and R. Kumar, ”Pattern Recognition Techniquesfor Acute Lymphoblastic Leukemia Detection,” *Biomedical SignalProcessing and Control*, vol. 69, pp. 102917, 2021.
[12] A. Abhishek, S. D. Deb, and K. Jha, ”Classification of Leukemia UsingFine-tuned VGG16 Deep Learning Model,” in *Proceedings of theInternationalConferenceonSmartComputingandElectronicEnterprise(ICSCEE)*, 2023, pp. 56–61.
[13] Q.UlAinandS.A.Shahzad,”DiagnosisofLeukemiaDiseaseThroughDeepLearningUsingMicroscopicImages,”*ComputersinBiologyandMedicine*, vol. 135, pp. 104538, 2021.
[14] E. Chen, H. Zhao, and L. Zhang, ”Real-Time Deep Learning-BasedDetection of Acute Lymphoblastic Leukemia Cells,” *Frontiers in On-cology*, vol. 12, 2022, Art. no. 895712.
[15] S. Rajeswari, B. Pradeep, and K. R. Lakshmi, ”Leukemia DetectionUsing Ensemble Learning Techniques,” *Journal of Healthcare Engi-neering*, vol. 2022, Article ID 3451734, 2022.
[16] F. Sakthiraj, ”Autonomous leukemia detection scheme based on hybridCNN model,” in *Artificial Intelligence in Medicine*, vol. 121, pp.102191, 2021.