[Paper Review] COVID-VIT: Classification of COVID-19 from CT chest images based on vision transformer models

Xiaohong Gao, Qian Yu+1 more|arXiv (Cornell University)|Jul 4, 2021

COVID-19 diagnosis using AIMedicine14 references43 citations

TL;DR

The paper compares vision transformer (ViT) and DenseNet for classifying COVID-19 from CT chest images, finding ViT superior in F1 on validation data.

ABSTRACT

This paper is responding to the MIA-COV19 challenge to classify COVID from non-COVID based on CT lung images. The COVID-19 virus has devastated the world in the last eighteen months by infecting more than 182 million people and causing over 3.9 million deaths. The overarching aim is to predict the diagnosis of the COVID-19 virus from chest radiographs, through the development of explainable vision transformer deep learning techniques, leading to population screening in a more rapid, accurate and transparent way. In this competition, there are 5381 three-dimensional (3D) datasets in total, including 1552 for training, 374 for evaluation and 3455 for testing. While most of the data volumes are in axial view, there are a number of subjects' data are in coronal or sagittal views with 1 or 2 slices are in axial view. Hence, while 3D data based classification is investigated, in this competition, 2D images remains the main focus. Two deep learning methods are studied, which are vision transformer (ViT) based on attention models and DenseNet that is built upon conventional convolutional neural network (CNN). Initial evaluation results based on validation datasets whereby the ground truth is known indicate that ViT performs better than DenseNet with F1 scores being 0.76 and 0.72 respectively. Codes are available at GitHub at .

Motivation & Objective

• Address rapid, explainable COVID-19 diagnosis from chest CT images for population screening.
• Compare ViT-based methods to traditional CNNs (DenseNet) for COVID vs non-COVID classification.
• Leverage 3D CT datasets while focusing on 2D slice-based classification for practicality and performance.

Proposed method

• Use ViT and DenseNet architectures to classify COVID-19 from chest CT slices.
• Evaluate performance on a mixed dataset with 3D volumes and slices (emphasizing 2D images).
• Report F1 scores on validation data to compare models (ViT vs DenseNet).
• Provide code availability on GitHub for reproducibility.

Experimental results

Research questions

• RQ1 Does vision transformer (ViT) outperform DenseNet for COVID-19 classification from CT chest images on validation data?
• RQ2 What are the comparative F1 scores of ViT and DenseNet in this task?
• RQ3 How effective is 2D slice-based CT data versus 3D data for this classification in the evaluated setup?
• RQ4 Is the approach explainable and suitable for population screening in practice?

Key findings

• ViT achieves higher F1 score (0.76) than DenseNet (0.72) on validation data.
• The study uses 5,381 total 3D datasets with 1,552 training, 374 evaluation, and 3,455 testing samples.
• Most data volumes are axial, but some subjects include coronal or sagittal views with 1–2 axial slices; 2D images remain the main focus.
• Code for the COVID-ViT approach is available on GitHub.