SELF-SUPERVISED VISION TRANSFORMERS FOR EARLY DISEASE DETECTION IN LOW-RESOURCE MEDICAL IMAGING ENVIRONMENTS FRAMEWORK
Qubahan Journal of Medical Sciences · 2025
DOI: 10.48161/qjms.v2n1a84 ↗Kurz gefasst
Fachartikel in der Zeitschrift Qubahan Journal of Medical Sciences, 2025. Verfasst von Anindya Bose und 2 weiteren Autor:innen. Die Arbeit nennt 23 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY-NC-ND 4.0).
Abstract (Originalsprache)
This study proposes a Self-Supervised Vision Transformer (ViT) framework for early disease detection in low-resource medical imaging environments. The research addresses major challenges in healthcare systems, including limited labeled medical imaging datasets, inadequate computational resources, and a shortage of medical experts. Traditional Convolutional Neural Network (CNN)-based models generally require large annotated datasets and extensive computational power, limiting their applicability in resource-constrained healthcare settings. To overcome these limitations, the proposed framework integrates self-supervised learning techniques with a Vision Transformer architecture to improve medical image classification performance using limited labeled data. The framework uses chest X-ray images for pneumonia detection and employs preprocessing methods, including normalization, augmentation, resizing, and denoising, to improve image quality and consistency. Self-supervised learning techniques, including contrastive learning and masked image modeling, are used to learn meaningful image representations from unlabeled data. The Vision Transformer architecture captures both local and global contextual information through self-attention mechanisms, enhancing feature extraction and disease representation. Experimental evaluation is conducted using metrics such as accuracy, precision, recall, F1-score, confusion matrix, and ROC-AUC, with a comparative analysis against the ResNet50 CNN model. The results demonstrate that the proposed Vision Transformer framework achieves strong disease-sensitive classification performance and excellent recall for pneumonia detection, while reducing dependency on large annotated datasets. Although the CNN model achieved higher overall classification accuracy, the Vision Transformer demonstrated superior contextual learning and disease-discrimination capabilities. The study highlights the potential of self-supervised Vision Transformers for scalable, efficient, and cost-effective AI-assisted medical image analysis in low-resource healthcare environments.
Zitieren
APA
Bose, A., Sooriamoorthy, D., & Murugasu, K. (2025). SELF-SUPERVISED VISION TRANSFORMERS FOR EARLY DISEASE DETECTION IN LOW-RESOURCE MEDICAL IMAGING ENVIRONMENTS FRAMEWORK. Qubahan Journal of Medical Sciences. https://doi.org/10.48161/qjms.v2n1a84
Vancouver
Bose A, Sooriamoorthy D, Murugasu K. SELF-SUPERVISED VISION TRANSFORMERS FOR EARLY DISEASE DETECTION IN LOW-RESOURCE MEDICAL IMAGING ENVIRONMENTS FRAMEWORK. Qubahan Journal of Medical Sciences. 2025. doi:10.48161/qjms.v2n1a84
BibTeX
@article{Bose2025,
author = {Bose, Anindya and Sooriamoorthy, Denesh and Murugasu, Kumaresu},
title = {SELF-SUPERVISED VISION TRANSFORMERS FOR EARLY DISEASE DETECTION IN LOW-RESOURCE MEDICAL IMAGING ENVIRONMENTS FRAMEWORK},
journal = {Qubahan Journal of Medical Sciences},
year = {2025},
doi = {10.48161/qjms.v2n1a84}
}
EUVIMED