Implementasi Convolutional Neural Network berbasis Transfer Learning untuk Klasifikasi Acute Lymphoblastic Leukemia

Abstract

Leukemia is a cancer that originates in human blood cells. The most common type of leukemia (97%), with an incidence of 4–4.5 cases per 100,000 children per year, is Acute Lymphoblastic Leukemia (ALL). This indicates that leukemia can progress rapidly and become fatal for the patient within a few months. Therefore, a supporting method is needed that can classify blood cells automatically, quickly, and accurately. This method is a Convolutional Neural Network (CNN) using the EfficientNet-B3 architecture as a pre-trained model or for transfer learning. This dataset consists of 3,527 blood cell images that have been preprocessed to a size of 224x224x3 and image enhancement has been applied. The images were trained on the pre-trained model and then combined with Global Average Pooling (GAP), Batch Normalization, Dense, and Softmax layers until the model could classify the images into the ALL or HEM classes. The results of the study show that the EfficientNet-B3 architecture is capable of classifying white blood cell images into the ALL and HEM classes through the transfer learning process. The best hyperparameter configuration for optimal results includes a learning rate of 0.0001 and the RMSProp optimizer. The model achieved the best training accuracy of 100% at epoch 30 and a batch size of 16, while the best testing accuracy was 96% at epoch 50 and a batch size of 16. Additionally, the precision, recall, and F1-score were 96%, 94%, and 95%, respectively.