Performance Evaluation of EfficientNetB3-Based Deep Learning Model for the Classification of Acute Lymphoblastic Leukemia and Normal Blood Cells

Acute Lymphoblastic Leukemia (ALL) is a rapidly progressing blood cancer that predominantly affects children and requires early and accurate diagnosis to improve patient survival rates. Traditional diagnostic methods rely heavily on manual examination of blood smear images by pathologists, which is not only time-consuming but also susceptible to human error and variability. To address this limitation, this study proposed an automated detection model based on deep learning, specifically employing the EfficientNetB3 convolutional neural network architecture. A publicly available dataset containing microscopic images of ALL and normal blood cells was used for training and evaluation. The images were preprocessed using normalization and augmentation techniques and resized to 300×300 pixels to align with the EfficientNetB3 input requirements. The model was trained using the Adam optimizer and monitored with EarlyStopping to prevent overfitting. Experimental results showed that the proposed model achieved an accuracy of 92.23%, precision of 92.75%, and recall of 95.57%, significantly outperforming conventional approaches such as Canberra distance, K-Nearest Neighbor, and ensemble CNN methods. In addition to the classification model, a web-based ALL detection system was developed to make the solution more accessible and user-friendly. The frontend was built using ReactJS, while the backend API, built with Flask, handles image input, model inference, and output delivery. The interface allows users to upload cell images, input patient names, and receive instant classification results along with confidence scores. This integrated system demonstrates a practical application of AI in medical diagnostics and holds potential for use in real-world, resource-limited clinical settings.