Deep Learning-based Blood Cell Classification using EfficientNetB3 Architecture
Aditya Kumar, Leema Nelson · 2025
The accurate classification of blood cell types is crucial for diagnosing a variety of hematological disorders. Yet, traditional methods relying on manual microscopy are time-consuming and prone to human error. Early detection and precise identification of blood cell abnormalities are essential for effective treatment and improved patient outcomes. This study addresses these challenges by developing and evaluating a deep learning model based on the EfficientNetB3 architecture for automated blood cell classification. The dataset from a Kaggle repository comprises 10,868 images categorized into six blood cell types: Eosinophil, Platelet, Erythroblast, Monocyte, Basophil, and Lymphocyte. To ensure a balanced and comprehensive evaluation, the dataset was split into training, validation, and testing subsets in an 80:10:10 ratio, enabling the model to learn effectively, fine-tune during training, and accurately assess performance. The model was rigorously tested and achieved an impressive overall test accuracy of 99.63%, with precision, recall, and F1-scores all reaching 1.00 across most classes, demonstrating its robustness and reliability in distinguishing between various blood cell types. These outstanding metrics validate the model's potential as a valuable tool in clinical diagnostics, where early and accurate detection of blood cell anomalies is vital. This research contributes significantly to the field of automated medical diagnostics, offering a promising solution for enhancing the accuracy and efficiency of blood cell classification.