Advancing WBC Classification: A Hybrid ConvNextV2-Swin Transformer Framework with R3GAN Data Balancing and CLAHE...

Area of Science:

• Medical Diagnostics
• Computational Biology
• Image Analysis

Background:

• Accurate white blood cell (WBC) classification is crucial for hematological diagnostics.
• Classifying rare cell types and handling imbalanced datasets present significant challenges.
• Existing methods struggle with data variability and limited sample sizes.

Purpose of the Study:

• To develop a novel hybrid framework for enhanced WBC classification.
• To address the challenges of rare cell types and imbalanced datasets in hematology.
• To improve the accuracy and efficiency of automated WBC classification systems.

Main Methods:

• A three-component hybrid framework integrating ConvNeXtV2-Swin Transformer for feature extraction.
• Utilizing Reinforced Reliable Robust Generative Adversarial Network (R3GAN) for intelligent minority class augmentation.
• Employing Contrast-Limited Adaptive Histogram Equalization (CLAHE) for adaptive image preprocessing.

Main Results:

• Achieved 99.1% accuracy on the challenging Raabin dataset, outperforming state-of-the-art methods by 2-10%.
• Demonstrated exceptional data efficiency, maintaining 94% accuracy with only 50% of the training data.
• Successfully mitigated class imbalance and preserved biological fidelity in generated samples.

Conclusions:

• The proposed framework offers a robust and accurate solution for WBC classification, particularly for rare cells and imbalanced data.
• The synergistic integration of advanced AI techniques and preprocessing provides a paradigm for clinical deployment.
• The framework's data efficiency makes it suitable for resource-constrained environments in hematological diagnostics.