CT Kernel Conversion for Quantitative Assessment in Chronic Obstructive Pulmonary Disease Using an Image‐To‐Image Schrödinger Bridge Model
International Journal of Imaging Systems and Technology · 2026-11 · 36 (6)
DOI: 10.1002/ima.70472 ↗Kurz gefasst
Fachartikel in der Zeitschrift International Journal of Imaging Systems and Technology, online erschienen am 05.11.2026. Verfasst von Doa Kim und 10 weiteren Autor:innen. Die Arbeit nennt 47 Literaturangaben. Gefördert von: Korea Health Industry Development Institute.
Themen: COPD
Abstract (Originalsprache)
Quantitative evaluation of chronic obstructive pulmonary disease (COPD) using computed tomography (CT) requires soft‐kernel reconstructions to accurately measure the emphysema index (EI) and airway wall thickness. However, many institutions archive only single‐kernel CT images, limiting retrospective analysis. This study assessed whether a diffusion‐based image translation model could generate reliable soft‐kernel equivalents from routinely stored sharp‐kernel CT images. In this retrospective study, paired soft‐ and sharp‐kernel CT scans were collected from a multi‐detector CT scanner. Three approaches—a convolutional neural network (CNN), a generative adversarial network (GAN), and the diffusion‐based Image‐to‐Image Schrödinger Bridge (I 2 SB)—were trained to convert sharp‐kernel to soft‐kernel images. Performance was evaluated using peak signal‐to‐noise ratio (PSNR), structural similarity index measure (SSIM), and root mean square error (RMSE); difference‐map analysis; Dice similarity coefficients (DSC) for low‐attenuation areas (LAA) and airway lumen; agreement in quantitative biomarkers (EI and standardized airway wall thickness, Pi10); and a blinded reader study involving radiologists and radiology residents. A total of 150 patients were included (mean age, 63.3 ± 13.7 years; 68.0% male). I 2 SB achieved higher PSNR and SSIM and lower RMSE than CNN and GAN (all p < 0.001). I 2 SB also produced higher DSC for LAA and airway lumen. For quantitative biomarkers, pix2pixHD showed a lower absolute error for EI, whereas I 2 SB showed a lower absolute error for Pi10. Reader evaluations also showed a preference for I 2 SB‐generated images. I 2 SB demonstrated strong overall performance for sharp‐to‐soft CT kernel conversion, although its relative advantage varied across quantitative biomarkers. These findings support the technical feasibility of I 2 SB‐based kernel conversion for retrospective quantitative COPD assessment.
Förderung
- Korea Health Industry Development Institute
Zitieren
APA
Kim, D., Choi, C., Lee, H. K., Nam, Y., Kwon, J., Kim, D. Y., Song, M., Choe, J., Ahn, Y., Lee, S. M., & Kim, N. (2026). CT Kernel Conversion for Quantitative Assessment in Chronic Obstructive Pulmonary Disease Using an Image‐To‐Image Schrödinger Bridge Model. International Journal of Imaging Systems and Technology, 36(6). https://doi.org/10.1002/ima.70472
Vancouver
Kim D, Choi C, Lee HK, Nam Y, Kwon J, Kim DY, et al. CT Kernel Conversion for Quantitative Assessment in Chronic Obstructive Pulmonary Disease Using an Image‐To‐Image Schrödinger Bridge Model. International Journal of Imaging Systems and Technology. 2026;36(6). doi:10.1002/ima.70472
BibTeX
@article{Kim2026,
author = {Kim, Doa and Choi, Changyong and Lee, Hyun Ki and Nam, Yoojin and Kwon, Jimin and Kim, Dong Yeong and Song, Mijeong and Choe, Jooae and Ahn, Yura and Lee, Sang Min and Kim, Namkug},
title = {CT Kernel Conversion for Quantitative Assessment in Chronic Obstructive Pulmonary Disease Using an Image‐To‐Image Schrödinger Bridge Model},
journal = {International Journal of Imaging Systems and Technology},
year = {2026},
volume = {36},
number = {6},
doi = {10.1002/ima.70472}
}
EUVIMED