AI-Enhanced Bone Marrow Diagnosis Through EfficientNetB5 Models

Classifying bone marrow images into specific categories is a complex task due to issues like uneven distribution of classes, high variability within the same category, and the need for precise diagnostic models. These challenges complicate the development of reliable automated systems for clinical use. Accurate identification of bone marrow images is crucial for diagnosing various blood diseases, yet multi-class classification faces obstacles such as the requirement for precise results, significant shape variations across classes, and imbalanced class distributions. This study explores the application of a modern convolutional neural network, EfficientNetB5, to enhance the classification of bone marrow images. Through the use of custom loss functions, advanced data augmentation techniques, and explainable tools, our approach achieves an accuracy of 96%. These improvements, which address class imbalance, increase dataset diversity, and enhance model interpretability, demonstrate that EfficientNetB5 can support more reliable and accurate diagnoses in medical contexts. By leveraging the strong architecture of EfficientNetB5, our method significantly boosts classification accuracy to 96%.