Blood Cell Cancer Classification Using the EfficientNetB3 Model: A Deep Learning Approach
Goldy Verma, Gotte Ranjith Kumar · 2025
Haematological diagnostics depend on accurate blood cell classification, which helps to detect and treat problems including leukemia and other blood diseases. This work shows a calibrated EfficientNetB3 model for six blood cell types: basophils, eosinophils, erythroblasts, lymphocytes, monocytes, and platelets. Comprising 17,092 annotated photos from the open-source Kaggle platform, a premium, well-balanced dataset, the model obtained an incredible 99% accuracy. Every image was categorized by expert clinical pathologists such that consistent ground truth labels required for effective training and validation could be assured. Among other performance indicators of the model in every class, precision, recall, and F1-scores were exceptional. For every kind of cell—especially eosinophils and platelets—the confusion matrix confirmed the consistency of accurate predictions. Highly accurate classification of even uncommon cell types including basophils and erythroblasts revealed the potential of the model to process different and demanding data. This work underlines how deep learning could be applied to automate blood cell classification, therefore reducing diagnosis errors and accelerating healthcare processes. The study also emphasizes the requirement of consistent datasets in improving AI-driven medical diagnoses, thereby encouraging reproducibility and scientific community cooperation to progress automated hematology analysis.