Challenges and Solutions in Multi-Class Bone Marrow Classification Using EfficientNetB5

In this research paper, we traverse the classification of bone marrow images by the application of deep learning techniques using the EfficientNetB5 model. Correct diagnosis and monitoring of certain hematological diseases depend on accurate identification of bone marrow cells. Though efficient, traditional hand techniques are time-consuming and prone to inter-observer variability. This work uses the modern Convolutional Neural Network EfficientNetB5 to automatically classify bone marrow cells into 21 unique groups. Source from respectable medical facilities and the Kaggle platform, the dataset guarantees variety and good quality. With an outstanding F1-score of 97% and general accuracy of 97%, our model regularly showed strong performance over all classes. Especially noteworthy in classes such "ART," "BAS," "BLA," "EBO," and "FGC were high F1-scores. With little misclassifications noted, precision, recall, and confusion matrix analysis highlighted the dependability of the model. Setting a new benchmark in bone marrow cell classification, this automated system not only improves diagnosis accuracy but also greatly lessens the burden on medical practitioners, therefore adding to the larger field of medical image analysis and automated diagnostics.