A Comparative Study of ViT-B16, DeiT, and ResNet50 for Peripheral Blood Cell Image Classification

This paper presents a comparative analysis of three advanced models, Vision Transformer (ViT-B16), Data-efficient Image Transformer (DeiT), and ResNet50, for the classification of peripheral blood cell (PBC) images. This study is focused on comparing the performance of these models, rather than introducing a novel approach. The study evaluates the models based on their accuracy, precision, recall, and F1-score on the PBC dataset. ViTB16 achieves an accuracy of 98.77%, significantly outperforming DeiT, which achieves an accuracy of 93.25%, and ResNet50, which achieves an accuracy of 73.01%. The results highlight the superior performance of ViT-B16 in terms of classification metrics, although DeiT demonstrates efficiency in data utilization and model complexity. ResNet50, while a robust convolutional neural network, shows lower accuracy in this specific task. This research provides valuable insights into the application of transformer-based models in medical image classification.