Implementasi Deep Learning dengan EfficientNet-B4 untuk Klasifikasi Acute Lymphoblastic Leukemia berdasarkan Citra Mikroskopis Sel Darah

Abstract

Acute Lymphoblastic Leukemia (ALL) is one of several types of blood cancer that is aggressive and progressive in nature, posing a significant risk of death if not identified and treated promptly. Accurate classification of leukemia stages is crucial in supporting precise medical diagnosis and enabling more effective treatment. With the rapid advancement of artificial intelligence technologies, particularly in the field of deep learning, image-based analysis of microscopic blood cells has become a promising approach to enhance diagnostic accuracy for this disease. This study aims to design a leukemia classification system based on deep learning using the EfficientNet-B4 architecture. The dataset used in this research consists of four main classes: Benign, Early, Pre, and Pro, each representing a specific stage in the progression of leukemia. The developed classification model demonstrated high performance, achieving a test accuracy of 96.25% a precision of 96.25%, a recall of 96.00%, and an F1-score of 96.25%, which indicates the model’s strong ability to differentiate between each category with a high level of accuracy. A comparison between EfficientNet-B4 and EfficientNet-B3 was also conducted, showing that EfficientNet-B4 performed better in classifying the stages of ALL. The system is visualized through a website interface that enables users to upload blood cell images and receive classification results quickly and efficiently. Based on the findings, the implementation of the EfficientNet-B4 architecture in the classification of ALL achieved high accuracy in testing. Furthermore, this research is expected to contribute to the early identification of ALL cases. The proposed system may serve as a valuable tool to assist medical professionals in classifying leukemia progression stages and providing more targeted and accurate diagnoses.