Implementation of Vision Transformers for Lung Abnormality Detection Using Low Dose CT Images

Alfred D., M. N. Deephak, D. Lakshmi

2025

Abstract

Detection of lung abnormalities originating from a variety of infections, inflammation, and environmental exposures in the patient needs high accuracy to help improve diagnostic efficiency by means of medical imaging. Regular CNNs have the weakest long-range dependencies, being very weak with almost zero receptive fields, which, therefore, induce many false positives and negatives. The future with ViTs is bright because the self-attention mechanism can extract local as well as global features from images and perform far better than regular CNNs. This work proposes a ViT-based model for the detection of lung abnormality in low-dose CT images. Most of the existing systems are prone to high classification error rates because of their poor quality towards understanding the context present in the images. The proposed ViT model bridges that gap by using a pre-trained architecture, and patch-based processing is used to focus more on the essential features of the image. We show how ViT's performance gets superseded by CNN for metrics through comparison. The overall goal behind the project would be facilitating the early detection of lung abnormalities, avoiding false results, and pushing the potential clinical uses with a dense, efficient solution for abnormality detection of the lung.

Paper Citation

in Harvard Style

D. A., Deephak M. and Lakshmi D. (2025). Implementation of Vision Transformers for Lung Abnormality Detection Using Low Dose CT Images. In Proceedings of the 1st International Conference on Research and Development in Information, Communication, and Computing Technologies - ICRDICCT`25; ISBN 978-989-758-777-1, SciTePress, pages 266-272. DOI: 10.5220/0013926400004919

in Bibtex Style

@conference{icrdicct`2525,
author={Alfred D. and M. Deephak and D. Lakshmi},
title={Implementation of Vision Transformers for Lung Abnormality Detection Using Low Dose CT Images},
booktitle={Proceedings of the 1st International Conference on Research and Development in Information, Communication, and Computing Technologies - ICRDICCT`25},
year={2025},
pages={266-272},
publisher={SciTePress},
organization={INSTICC},
doi={10.5220/0013926400004919},
isbn={978-989-758-777-1},
}

in EndNote Style

TY - CONF

JO - Proceedings of the 1st International Conference on Research and Development in Information, Communication, and Computing Technologies - ICRDICCT`25
TI - Implementation of Vision Transformers for Lung Abnormality Detection Using Low Dose CT Images
SN - 978-989-758-777-1
AU - D. A.
AU - Deephak M.
AU - Lakshmi D.
PY - 2025
SP - 266
EP - 272
DO - 10.5220/0013926400004919
PB - SciTePress