Bone Marrow Classification Using Hematologic Malignancies Dataset

Morphological analysis of bone marrow cells is considered the most reliable method for diagnosing leukemia. However, due to diverse cell appearances, it requires extensive expertise and patience. An automated diagnostic system combining image analysis and pattern recognition technology is needed to reduce workload, minimize errors, and enhance overall work efficiency. This paper proposes a deep learning-based strategy for classifying bone marrow samples using ResNet50, DenseNet121, and EfficientNet models. After training and assessment on a large dataset, DenseNet121 was the most accurate (98%), followed by ResNet50 (97%). EfficientNet, with a lower accuracy of 69%, still provided insightful information. These results demonstrate the effectiveness of deep learning models in classifying bone marrow, paving the way for improved leukemia diagnosis and treatment.