Comparison of CNN Architecture for White Blood Cells Image Classification

Deep learning is a development of machine learning that has obtained good performance in image segmentation, prediction, classification, and machine translation. CNN has a variety of architectures that have undergone various changes and improvements. One area of medicine that requires automatic classification is white blood cell classification. The traditional blood cell classification process has many shortcomings, one of which is inefficient and takes a long time. In addition, some conventional methods can also damage the blood cell sample. This study aims to perform a comparative analysis of the performance of CNN architectures such as AlexNet, VGG19, VGG16, and ResNet50, improve accuracy performance and classify white blood cells using blood cell images. The research process involves creating and training models for each architecture using the same input data. The results showed that the model built with the ResNet50 architecture required the longest training time of 3391 seconds but produced the best accuracy in white blood cell classification of 99%, with recall, precision, and 99% f1-score values. Meanwhile, models with AlexNet, VGG19, and VGG16 architectures only obtained an accuracy of 98%, 94%, and 95%, respectively.