Clinical Report: Ensemble Learning for Automated Classification of Bone Marrow Cells
Overview
This study presents an ensemble-learning framework that significantly enhances the classification accuracy of bone marrow cells, achieving 96% accuracy with robust generalization across various imaging conditions. The model's interpretability is improved through explainable AI methods, supporting its clinical applicability in hematological diagnostics.
Background
Bone marrow cell classification is critical for diagnosing hematological disorders, yet it faces challenges due to morphological similarities and imaging artifacts. Traditional manual classification methods are slow and prone to interobserver variability, highlighting the need for automated solutions. Deep learning models, particularly ensemble approaches, offer a promising avenue to improve diagnostic accuracy and efficiency in this field.
Data Highlights
Ensemble Strategy
Classification Accuracy
Boosting
96%
Key Findings
The ensemble model outperformed individual models in classification accuracy and interpretability stability.
Boosting strategy achieved the highest classification accuracy of 96%.
External validation confirmed robust performance across independent datasets.
Explainable AI methods demonstrated that the model focused on relevant morphological features.
The framework is computationally efficient, making it suitable for clinical applications.
Clinical Implications
The MobileNetV3–ResNet18 ensemble framework can enhance the accuracy and reliability of bone marrow cell classification, potentially reducing diagnostic time and interobserver variability. Its interpretability supports clinicians in understanding AI-driven decisions, fostering trust in automated systems.
Conclusion
The proposed ensemble learning framework represents a significant advancement in automated bone marrow cell classification, with strong implications for improving diagnostic processes in hematology.
