LeukoDetect: Leukemia Detection Using Deep Learning for Medical Image Analysis

Abstract

The bloodstream cancer known as Acute Lymphoblastic Leukaemia (ALL) progresses rapidly because it develops through the uncontrolled growth of immature lymphoid cells, which requires doctors to diagnose it correctly and treat it with appropriate methods for better patient outcomes. The traditional diagnostic process which relies on experts to examine peripheral blood smears through microscopy requires a considerable amount of time and shows different results depending on the examiner while becoming more difficult to execute because there is a worldwide deficit of qualified haematopathologists. The research introduces LeukoDetect which serves as a computer-aided diagnostic system that uses deep learning technology to automatically sort blood smear images into two distinct categories: malignant ALL and normal Hematogone (HEM) samples. The transfer learning approach uses DenseNet121 which is a densely connected convolutional network that has been pretrained on ImageNet to achieve efficient feature sharing and strong gradient propagation capabilities on the medical dataset from this specific domain. The researchers conducted their experiments using the publicly accessible C-NMC 2019 benchmark which includes 10,661 segmented lymphocyte images. The system applies complete preprocessing procedures which include image resizing and per-channel normalisation and stochastic image enhancement before the fine-tuning stage. LeukoDetect achieves a classification accuracy of 91% and a precision score of 1.00 and a recall score of 0.91 and an F1-score of 0.95 on the held-out test partition while surpassing six competing baselines which include standard CNN and MobileNetV2 and ResNet18 and Random Forest and SVM. The results demonstrate that the proposed system can serve as an effective clinical tool which helps doctors make decisions about early leukaemia detection with minimal resource requirements.

Introduction

Leukaemia, particularly Acute Lymphoblastic Leukaemia (ALL), is a fast-progressing and common childhood blood cancer that requires early and accurate diagnosis. Traditional diagnosis methods using manual microscopic blood smear analysis are time-consuming, labor-intensive, and prone to limitations in scalability and resource availability.

To address these challenges, the study introduces LeukoDetect, an automated leukemia detection system based on deep learning. The system uses transfer learning with DenseNet121 to classify blood smear images as malignant (ALL) or normal (HEM). It includes preprocessing, data augmentation, and handling of class imbalance using the C-NMC 2019 dataset.

A literature review compares different approaches such as SVM, Random Forest, CNNs, ResNet18, MobileNetV2, and DenseNet121, showing that traditional machine learning methods perform poorly due to reliance on handcrafted features, while deep learning models perform better but still face issues like overfitting, false positives, or low recall. DenseNet121 emerges as the most balanced and effective model among them.

The methodology involves a full pipeline: image acquisition, preprocessing, augmentation, feature extraction via DenseNet121, classification, training, and evaluation. The dataset contains 10,661 images with class imbalance between ALL and HEM samples, which is addressed during training.

Conclusion

The research study described in this document introduces LeukoDetect which functions as a fully automated system to identify leukemia through its use of DenseNet121 transfer learn-ing technology. The proposed system achieves an accuracy of 91%, precision of 1.00, recall of 0.91, and an F1-score of 0.95 which surpasses the performance of all baseline methods that were tested including CNN, MobileNetV2, ResNet18, Random Forest, and SVM. The system employs a strong preprocess-ing pipeline which combines transfer learning with fine-tuning to successfully handle the issues that arise from class imbal-ance and insufficient training data and the morphological resem-blances between ALL and HEM cells. LeukoDetect shows high potential to function as a dependable computer-aided diagnos-tic system which operates with low computational demands to assist in the early detection of leukemia through its support of clinicians who work in areas with poor access to haematopathol-ogy professionals. The system achieves automated analysis of peripheral blood smear images which enables faster diagnosis times while decreasing the differences in assessment results between different observers. The upcoming research will investigate methods for classify-ing ALL subtypes which will utilize Grad-CAM for creating interpretable results and enable testing across various datasets while establishing capabilities for clinical environments with limited resources.

References

[1] S. Mourya, S. Kant, P. Kumar, A. Gupta, and R. Gupta, “ALL Challenge Dataset of ISBI 2019 (C-NMC 2019),” The Cancer Imaging Archive, 2019. [2] R. Gupta, S. Gehlot, and A. Gupta, “C-NMC: B-lineage Acute Lym-phoblastic Leukaemia: A Blood Cancer Dataset,” Medical Engineering & Physics, vol. 103, p. 103793, 2022. [3] Y. LeCun, Y. Bengio, and G. Hinton, “Deep Learning,” Nature, vol. 521, no. 7553, pp. 436–444, 2015. [4] J. Pan and Q. Yang, “A Survey on Transfer Learning,” IEEE Transactions on Knowledge and Data Engineering, vol. 22, no. 10, pp. 1345–1359, 2010. [5] G. Huang, Z. Liu, L. van der Maaten, and K. Q. Weinberger, “Densely Con-nected Convolutional Networks,” in Proc. IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pp. 2261–2269, 2017. [6] K. Dese, H. Raj, T. Ayana, T. Sahile, and W. Habtewold, “Accurate Machine-Learning-Based Classification of Leukemia from Blood Smear Images,” Clinical Lymphoma, Myeloma and Leukemia, vol. 21, no. 11, pp. e903–e914, 2021. [7] N. Anand, S. Prabhu, and K. Viswanathan, “A Systematic Review of Ma-chine and Deep Learning Techniques for Acute Lymphoblastic Leukemia Diagnosis,” Biomedical Signal Processing and Control, vol. 90, p. 105864, 2024. [8] K. He, X. Zhang, S. Ren, and J. Sun, “Deep Residual Learning for Image Recognition,” in Proc. IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pp. 770–778, 2016. [9] S. Goswami et al., “Heterogeneity Loss to Handle Intersubject Variability in Cancer,” arXiv preprint arXiv:2003.03295, 2020. [10] A. Ahmed et al., “Leukemia Detection Based on Microscopic Blood Smear Images Using Deep Learning,” arXiv preprint arXiv:2301.03367, 2022. [11] M. Sandler, A. Howard, M. Zhu, A. Zhmoginov, and L.-C. Chen, “Mo-bileNetV2: Inverted Residuals and Linear Bottlenecks,” in Proc. IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pp. 4510–4520, 2018. [12] N. K. Honnalgere and G. Nayak, “Classification of Normal Versus Malig-nant Cells in B-ALL White Blood Cancer Microscopic Images,” in ISBI 2019 C-NMC Challenge: Classification in Cancer Cell Imaging, Springer, pp. 1–12, 2019. [13] J. Prellberg and O. Kramer, “Acute Lymphoblastic Leukemia Classification from Microscopic Images Using Convolutional Neural Networks,” in Proc. ISBI 2019 C-NMC Challenge, Springer, pp. 53–61, 2019. [14] O. Russakovsky et al., “ImageNet Large Scale Visual Recognition Chal-lenge,” International Journal of Computer Vision, vol. 115, no. 3, pp. 211–252, 2015. [15] C. Shorten and T. M. Khoshgoftaar, “A Survey on Image Data Augmen-tation for Deep Learning,” Journal of Big Data, vol. 6, no. 1, pp. 1–48, 2019. [16] R. R. Selvaraju, M. Cogswell, A. Das, R. Vedantam, D. Parikh, and D Batra, “Grad-CAM: Visual Explanations from Deep Networks via Gradient-Based Localization,” in Proc. IEEE International Conference on Computer Vision (ICCV), pp. 618–626, 2017.

Copyright

Copyright © 2026 Purva Sachin Sonaje, Vaishnavi Sunil Borase, Swamini Dipak Patil, Akash Narendra Shinde, P. D. Lanjewar. 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.