To develop an ensemble-learning framework for improved classification of bone marrow cells, specifically addressing challenges such as morphological overlap, class imbalance, and imaging artifacts in automated classification.
Key Findings:
Boosting achieved the highest classification accuracy of 96%.
The ensemble model outperformed individual models in classification accuracy and demonstrated greater stability in interpretability.
XAI analysis showed the model focused on relevant morphological features, supporting clinical plausibility.
Interpretation:
The MobileNetV3–ResNet18 ensemble framework provides accurate, computationally efficient, and interpretable classification of bone marrow cells, with significant implications for clinical practice.
Limitations:
Potential limitations include the reliance on high-quality imaging and the need for further validation across diverse clinical settings.
Conclusion:
The approach has strong potential to support AI-assisted hematological diagnostics, reduce diagnostic time, and minimize interobserver variability.
