Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Enhanced Leukemic Cell Classification Using Attention-Based CNN and Data Augmentation

Douglas Costa Braga, Daniel Oliveira Dantas

TL;DR

The study tackles automated leukemic cell classification despite inter-observer variability by proposing an attention-based CNN that combines EfficientNetV2-B3 with Squeeze-and-Excitation attention. It emphasizes reproducible pipelines, aggressive data augmentation, focal loss for imbalance, and a patient-wise evaluation protocol on the C-NMC 2019 dataset. The model achieves state-of-the-art performance with a test F1-score of $97.89\%$ and accuracy of $97.89\%$, using $15.2$M parameters (≈$89\%$ fewer than VGG16) and robust Monte Carlo validation ($p<0.001$) that confirms significance and generalization, complemented by interpretable attention visualizations. This work demonstrates that modern attention-based architectures can deliver high-accuracy, efficient leukemic cell classification suitable for clinical deployment while maintaining reproducibility and interpretability.

Abstract

We present a reproducible deep learning pipeline for leukemic cell classification, focusing on system architecture, experimental robustness, and software design choices for medical image analysis. Acute lymphoblastic leukemia (ALL) is the most common childhood cancer, requiring expert microscopic diagnosis that suffers from inter-observer variability and time constraints. The proposed system integrates an attention-based convolutional neural network combining EfficientNetV2-B3 with Squeeze-and-Excitation mechanisms for automated ALL cell classification. Our approach employs comprehensive data augmentation, focal loss for class imbalance, and patient-wise data splitting to ensure robust and reproducible evaluation. On the C-NMC 2019 dataset (12,528 original images from 62 patients), the system achieves a 97.89% F1-score and 97.89% accuracy on the test set, with statistical validation through 100-iteration Monte Carlo experiments confirming significant improvements (p < 0.001) over baseline methods. The proposed pipeline outperforms existing approaches by up to 4.67% while using 89% fewer parameters than VGG16 (15.2M vs. 138M). The attention mechanism provides interpretable visualizations of diagnostically relevant cellular features, demonstrating that modern attention-based architectures can improve leukemic cell classification while maintaining computational efficiency suitable for clinical deployment.

Enhanced Leukemic Cell Classification Using Attention-Based CNN and Data Augmentation

TL;DR

The study tackles automated leukemic cell classification despite inter-observer variability by proposing an attention-based CNN that combines EfficientNetV2-B3 with Squeeze-and-Excitation attention. It emphasizes reproducible pipelines, aggressive data augmentation, focal loss for imbalance, and a patient-wise evaluation protocol on the C-NMC 2019 dataset. The model achieves state-of-the-art performance with a test F1-score of and accuracy of , using M parameters (≈ fewer than VGG16) and robust Monte Carlo validation () that confirms significance and generalization, complemented by interpretable attention visualizations. This work demonstrates that modern attention-based architectures can deliver high-accuracy, efficient leukemic cell classification suitable for clinical deployment while maintaining reproducibility and interpretability.

Abstract

We present a reproducible deep learning pipeline for leukemic cell classification, focusing on system architecture, experimental robustness, and software design choices for medical image analysis. Acute lymphoblastic leukemia (ALL) is the most common childhood cancer, requiring expert microscopic diagnosis that suffers from inter-observer variability and time constraints. The proposed system integrates an attention-based convolutional neural network combining EfficientNetV2-B3 with Squeeze-and-Excitation mechanisms for automated ALL cell classification. Our approach employs comprehensive data augmentation, focal loss for class imbalance, and patient-wise data splitting to ensure robust and reproducible evaluation. On the C-NMC 2019 dataset (12,528 original images from 62 patients), the system achieves a 97.89% F1-score and 97.89% accuracy on the test set, with statistical validation through 100-iteration Monte Carlo experiments confirming significant improvements (p < 0.001) over baseline methods. The proposed pipeline outperforms existing approaches by up to 4.67% while using 89% fewer parameters than VGG16 (15.2M vs. 138M). The attention mechanism provides interpretable visualizations of diagnostically relevant cellular features, demonstrating that modern attention-based architectures can improve leukemic cell classification while maintaining computational efficiency suitable for clinical deployment.
Paper Structure (21 sections, 1 equation, 3 figures, 4 tables)

This paper contains 21 sections, 1 equation, 3 figures, 4 tables.

Table of Contents

  1. INTRODUCTION
  2. METHODOLOGY
  3. Dataset
  4. Preprocessing
  5. Data Augmentation
  6. Model Training
  7. Backbone Network
  8. Attention Mechanism
  9. Classification Head
  10. Loss Function
  11. Training Strategy
  12. Evaluation Metrics
  13. Statistical Validation
  14. Implementation Details
  15. RESULTS AND DISCUSSION
  16. ...and 6 more sections

Figures (3)

  • Figure 1: Overview of the proposed methodology: (a) Data preprocessing and augmentation, (b) Attention-based CNN architecture, and (c) Classification output.
  • Figure 2: Training and Validation curves showing (a) Loss, (b) Accuracy, and (c) F1-score progression. Best Validation performance achieved at epoch 10 with F1-score of 98.37%.
  • Figure 3: Attention maps visualization: (a,c) Original images of malignant and healthy cells, (b,d) Corresponding attention maps highlighting diagnostically relevant regions.