Manuscript received July 22, 2025; revised September 3, 2025; accepted October 24, 2025; published December 19, 2025.

Abstract—Blood cancer is a type of cancer which affected many people and their lives have flipped totally while battling the cancer. Most common type of blood cancer is Acute Lymphocytic Leukemia (ALL), Acute Myeloid Leukemia (AML) and Multiple Myeloma (MM). One of the primary tests done by pathologist to confirm the diagnosis is microscopic blood smear analysis, which requires human interpretation and consumes a lot of time. To overcome these problems, the proposed study helps the pathologist with a distinctive and a novel automated neural network mechanism, called HemoNet, which is a Convolutional Neural Networks (CNNs) architecture with a Convolutional Block Attention Module (CBAM) mechanism, incorporated to correctly identify the type of cancer within less time. This study poses a helping hand rather than completely taking over the human interpretation. The proposed work has been performed over ALL, AML and MM blood smear image dataset. Data transformation was performed to overcome the overfitting, allowing the model to learn the features with different characteristics. The baseline model ‘DeepLeukNet’ which when trained on the dataset, obtained an accuracy of 99.56% with a validation loss of 0.4. Whereas the proposed HemoNet model which has CBAM module, obtained an accuracy of 99.56% with a validation loss of 0.34, and about 79% faster.

Keywords—deep learning, Convolutional Neural Networks (CNN), attention mechanisms, optimization, blood smear analysis, Convolutional Block Attention Module (CBAM), image analysis

Cite: Mageshwari V., Jana U. Sagar, and Ashok K. M., "Leukemia Cancer Classification via Attention- Guided Deep Networks on Microscopic Smear Images," Journal of Image and Graphics, Vol. 13, No. 6, pp. 709-721, 2025.

Copyright © 2025 by the authors. This is an open access article distributed under the Creative Commons Attribution License (CC-BY-4.0), which permits use, distribution and reproduction in any medium, provided that the article is properly cited, the use is non-commercial and no modifications or adaptations are made.