AI Meets Hematology: Enhanced Bone Marrow Diagnostics with EfficientNetB5

Classifying bone marrow images into various categories poses significant challenges due to issues such as uneven distribution of classes, substantial variability within the same class, and the necessity for precise and reliable diagnostic models. These difficulties hinder the development of dependable automated systems for clinical diagnosis. Accurate identification of bone marrow images is crucial for detecting various blood disorders. Multi-class classification faces several hurdles, including the demand for precise outcomes, considerable shape differences among classes, and class imbalances. This study explores the application of a state-of-the-art convolutional neural network, EfficientNetB5, for classifying bone marrow images. Our approach achieves an impressive 96% accuracy by employing specialized loss functions, innovative data augmentation techniques, and transparent interpretability tools. The findings demonstrate that EfficientNetB5 significantly enhances the reliability and accuracy of diagnoses in medical contexts. To tackle these challenges, we implemented tailored loss functions to address class imbalance, advanced data enhancement strategies to diversify our dataset, and explainable methodologies to ensure the clarity and trustworthiness of our model's predictions. By leveraging the robust architecture of EfficientNetB5, our method achieves a remarkable 96% classification accuracy, marking a substantial advancement in this field.