Progressive Refinement Autoencoder for Low‐Dose Computed Tomography Enhancement

Demir A., Ashames M. M. A., Gerek O. N., Ergin S., Fidan M., Koc M., ...Daha Fazla

INTERNATIONAL JOURNAL OF IMAGING SYSTEMS AND TECHNOLOGY, ss.1-30, 2026 (SCI-Expanded, Scopus)

  • Yayın Türü: Makale / Tam Makale
  • Basım Tarihi: 2026
  • Doi Numarası: 10.1002/ima.70438
  • Dergi Adı: INTERNATIONAL JOURNAL OF IMAGING SYSTEMS AND TECHNOLOGY
  • Derginin Tarandığı İndeksler: Applied Science & Technology Source, Academic Search Ultimate (EBSCO), Engineering Source (EBSCO), Scopus, Technology Collection (ProQuest), Aerospace Database, Science Citation Index Expanded (SCI-EXPANDED), Compendex, INSPEC
  • Sayfa Sayıları: ss.1-30
  • Eskişehir Osmangazi Üniversitesi Adresli: Evet

Özet

The use of ionizing radiation in diagnostic imaging is a common practice worldwide. However, the imaging process itself carries relative risks. Therefore, it is recommended to employ the lowest possible dose of ionizing radiation, especially in computed tomography (CT) imaging, where a series of X-ray scans are utilized to reconstruct images of body tissue sections. A common strategy for radiation dose reduction in CT imaging is known as the quarter-dose technique, which reduces the X-ray dose but can cause noise and loss of image sharpness. Since CT image reconstruction from directional X-rays is a nonlinear process, it is analytically difficult to correct the effects of dose reduction on image quality. Recent and popular deep-learning approaches provide an intriguing possibility of low-dose CT image enhancement. This study aims to develop a U-Net-based deep-learning method with decoder-side refinement modules for enhancing quarter-dose CT images while preserving diagnostically relevant image details. A novel autoencoder approach is proposed that preserves the conventional encoding backbone of U-Net while integrating lightweight refinement modules exclusively within the decoding path. The refinement module consists of residual blocks for both coarse and fine resolutions, and the proposed method progressively updates feature representations to enhance fine details. The proposed architecture is evaluated through experiments conducted on the Low Dose CT Grand Challenge dataset by comparing it with well-known architectures. The dataset consists of quarter-dose and full-dose CT slice images from 10 patients. The model is evaluated using both an initial fixed-interval test set and an additional patient-wise stratified random test set comprising 300 paired slices, and the remaining images are divided into training (90%) and validation (10%) sets. Moreover, blind tests conducted by practicing radiologists demonstrate that the enhanced quarter-dose CT images received higher radiologist-rated diagnostic acceptability scores than the original quarter-dose images and are often preferred in subjective comparison with the full-dose CT images. The proposed progressive refinement U-Net provides a decoder-side refinement strategy for quarter-dose CT enhancement and achieves competitive quantitative performance. Beyond demonstrating successful image enhancement, the study also highlights the importance of radiologist-rated diagnostic acceptability and structured subjective comparison when assessing low-dose CT restoration methods.