Comparative Study of Convolutional Neural Network Architecture in Lymphoma Detection

Abstract

In this study, we propose an automatic classification of three common types of lymphoma: (1) lymphoma, (2) benign lesion, and (3) carcinoma using lymphoma cell images magnified by 100x and by 400x. A comparative study was performed to find the best architecture to classify lymphoma cell images using the Keras library in Tensorflow. The architectures used in this study are ResNet50, MobileNetV1, and VGG16. Based on the accuracy of lymphoma classification for each architecture, the MobileNet model had the highest accuracy in all three classes at both 100x and 400x magnification levels, which suggests that MobileNet is the best model for lymphoma cell classification. This study can be later used as the base argument in modifying the MobileNet architecture further to get more accurate results in future similar studies.