A Hybrid Super-Resolution Method for Medical Images Using Transfer Learning
Journal of Biomedical Photonics & Engineering · 30.12.2025 · 11 (4) · 040304
DOI: 10.18287/jbpe25.11.040304 ↗Kurz gefasst
Fachartikel in der Zeitschrift Journal of Biomedical Photonics & Engineering, online erschienen am 30.12.2025. Verfasst von Amir Hossein Jalalzadeh und 1 weiteren Autor:innen. Die Arbeit wurde bisher 1-mal zitiert (Crossref).
Abstract (Originalsprache)
High-quality medical imaging is essential for accurate diagnosis and effective treatment planning. However, limitations such as noise, artifacts, high acquisition costs, and patient safety concerns often restrict the ability to obtain high-resolution (HR) scans. Super-resolution (SR) techniques offer a promising alternative by reconstructing HR images from low-resolution (LR) inputs, thereby enhancing image quality without additional scanning. This capability is particularly important in clinical contexts, where improving image resolution directly impacts diagnostic confidence and therapeutic decision-making. In this work, we propose a hybrid SR framework that integrates transfer learning with a generative adversarial network (RESRGAN) and low-rank total variation (LRTV) regularization. This combination leverages both local neighborhood features and global contextual information to recover fine structural details while preserving overall image fidelity. The approach was evaluated on three brain imaging modalities – magnetic resonance imaging (MRI), computed tomography (CT), and ultrasound – using volumetric datasets acquired at different time intervals. Quantitative experiments show that the proposed method consistently outperforms bicubic interpolation, achieving an average peak signal-to-noise ratio (PSNR) of 40.05 dB and a structural similarity index (SSIM) of 0.9808. Notably, ultrasound images exhibited the most substantial improvement. These findings highlight the potential of the proposed framework as a robust tool for enhancing medical image quality and supporting more reliable clinical decision-making.
Zitieren
APA
Jalalzadeh, A. H., & Kamangar, M. H. (2025). A Hybrid Super-Resolution Method for Medical Images Using Transfer Learning. Journal of Biomedical Photonics & Engineering, 11(4), 040304. https://doi.org/10.18287/jbpe25.11.040304
Vancouver
Jalalzadeh AH, Kamangar MH. A Hybrid Super-Resolution Method for Medical Images Using Transfer Learning. Journal of Biomedical Photonics & Engineering. 2025;11(4):040304. doi:10.18287/jbpe25.11.040304
BibTeX
@article{Jalalzadeh2025,
author = {Jalalzadeh, Amir Hossein and Kamangar, Mehrdad Hashemi},
title = {A Hybrid Super-Resolution Method for Medical Images Using Transfer Learning},
journal = {Journal of Biomedical Photonics & Engineering},
year = {2025},
volume = {11},
number = {4},
pages = {040304},
doi = {10.18287/jbpe25.11.040304}
}
EUVIMED