Loading article...
Loading article...
Stream/Area:
-
Subject:
-
Paper ID:
IJSREM64762
Volume:
10 |
6 |
Jun 2026
Pages:
1-8
Publication Frequency:
Monthly (12 issues per year)
License:International Journal of Scientific Research in Engineering and Management follows the Creative Commons Attribution 4.0 (CC BY 4.0) License, allowing free use and distribution with proper attribution.
Computer Science And Engineering
University Of Visvesvaraya College Of Engineering
Computer Science And Engineering
University Of Visvesvaraya College Of Engineering
Computer Science And Engineering
University Of Visvesvaraya College Of Engineering
Computer Science And Engineering
University Of Visvesvaraya College Of Engineering
Pneumonia remains a leading cause of global mortality, particularly among children under five, and is typically diagnosed through chest X-ray (CXR) imaging. Deep learning (DL) offers a scalable alternative to expert-driven diagnosis; however, existing approaches suffer from domain mismatch between natural image pre-training and medical imaging, as well as insufficient input quality preprocessing. This paper proposes EDAViT (Enhanced Domain-Adapted Vision Transformer), an extension of the DAViT framework that introduces a dedicated image quality enhancement stage — comprising Gaussian noise removal — applied before the standard preprocessing pipeline, and incorporates additional publicly sourced chest X-ray training data totalling approximately 5000 images. The core hybrid architecture retaining the DINOv2-large Vision Transformer (ViT) for global feature extraction, a shallow Convolutional Neural Network (CNN) for local feature refinement, and domain adaptation via the ChestX-ray8 (CXR8) dataset is preserved. Evaluated on the standard Kermany et al. benchmark, EDAViT achieves a 97.31% F1-score and 96.14% AUC for pneumonia detection, and an 82.14% F1-score and 85.19% AUC for pneumonia type classification, surpassing the original DAViT baseline and twelve competing methods. An ablation study confirms the contribution of each component, and Grad-CAM visualisations provide clinically interpretable heatmaps that highlight infection regions in bacterial and viral pneumonia cases.
KEYWORDS
This paper presented EDAViT, an enhanced version of the DAViT framework for automated pneumonia detection and type classification from chest X-ray images. By introducing a Gaussian denoising stage prior to the standard preprocessing pipeline and augmenting the training dataset with additional publicly sourced CXR images, EDAViT achieves consistent improvements over the DAViT baseline across all evaluation metrics. On the standard Kermany et al. benchmark, EDAViT attains a 97.31% F1-score and 96.14% AUC for pneumonia detection, and an 82.14% F1-score and 85.19% AUC for pneumonia type classification, outperforming all twelve baseline methods compared. The ablation study confirms the contribution of each component — pre-training, denoising, CNN, and domain adaptation — and Grad-CAM visualisations provide clinically interpretable activation maps consistent with known radiological characteristics of bacterial and viral pneumonia. EDAViT demonstrates that targeted input quality enhancement, even without architectural changes, can yield measurable performance gains in transformer-based medical image classification.
[1] Amer. Lung Assoc. (2024). Learn About Pneumonia. [Online]. Available: https://www.lung.org/lung-health-diseases/lung-disease-lookup/pneumonia/learn-about-pneumonia [2] Centre for Tropical Med. Global Health. (2023). Celebrating World Pneumonia Day 2023. [Online]. Available: https://www.tropicalmedicine.ox.ac.uk/news/celebrating-world-pneumonia-day-2023 [3] D. Kermany et al., "Identifying medical diagnoses and treatable diseases by image-based deep learning," Cell, vol. 172, no. 5, pp. 1122-1131, Feb. 2018. [4] M. Fu, C. Tantithamthavorn, and T. Le, "DAViT: A Domain-Adapted Vision Transformer for Automated Pneumonia Detection and Explanation Using Chest X-Ray Images," IEEE Access, vol. 13, 2025, doi: 10.1109/ACCESS.2025.3579314. [5] R. Kailasam and S. Balasubramanian, "Deep learning for pneumonia detection: A combined CNN and YOLO approach," Human-Centric Intell. Syst., vol. 5, no. 1, pp. 44-62, Mar. 2025. [6] J. Deng, W. Dong, R. Socher, L.-J. Li, K. Li, and L. Fei-Fei, "ImageNet: A large-scale hierarchical image database," in Proc. IEEE CVPR, Jun. 2009, pp. 248-255. [7] A. Dosovitskiy et al., "An image is worth 16x16 words: Transformers for image recognition at scale," arXiv:2010.11929, 2020. [8] Ultralytics. (2023). Explore Ultralytics YOLOv8. [Online]. Available: https://docs.ultralytics.com/models/yolov8/ [9] Ultralytics. (2024). Ultralytics YOLO11. [Online]. Available: https://docs.ultralytics.com/models/yolo11/ [10] K. He, X. Zhang, S. Ren, and J. Sun, "Deep residual learning for image recognition," in Proc. IEEE CVPR, Jun. 2016, pp. 770-778.
Publication Date
6/10/2026
Peer Review Type
Double-Blind Peer Review
Review Rounds: 2
Documents
Article Views
0
Total views since publication
PDF Downloads
5
Total downloads since publication
Geographic Reach
5
Digital Preservation: This article is permanently archived on Zenodo and complies with all data preservation standards across major indexing platforms.
Plagiarism Screening
This manuscript has been checked for plagiarism using TurnitinPlagiarismChecker software.
Conflict of Interest
The authors declare no conflict of interest related to this research work.
Funding Information
No funding information available for this research.
Related research in the same subject area
Loading similar articles...