WEKO3
アイテム
Radiomics reproducibility analysis of generative adversarial network-based super-resolution for accelerating brain MRI in T2*-weighted images
https://repo.qst.go.jp/records/2001684
https://repo.qst.go.jp/records/2001684e08812af-d849-4d1d-9a09-55aca103c1c5
| アイテムタイプ | 学術雑誌論文 / Journal Article(1) | |||||||||||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 公開日 | 2025-07-16 | |||||||||||||||||||||||
| タイトル | ||||||||||||||||||||||||
| タイトル | Radiomics reproducibility analysis of generative adversarial network-based super-resolution for accelerating brain MRI in T2*-weighted images | |||||||||||||||||||||||
| 言語 | en | |||||||||||||||||||||||
| 言語 | ||||||||||||||||||||||||
| 言語 | eng | |||||||||||||||||||||||
| 資源タイプ | ||||||||||||||||||||||||
| 資源タイプ識別子 | http://purl.org/coar/resource_type/c_6501 | |||||||||||||||||||||||
| 資源タイプ | journal article | |||||||||||||||||||||||
| 著者 |
Umehara Kensuke
× Umehara Kensuke
× Tatsuya Nishii
× Ota Junko
× Hiroki Nakajima
× Ryogo Enoki
× Yasutoshi Ohta
× Tetsuya Fukuda
× Ohba Hisateru
× Obata Takayuki
|
|||||||||||||||||||||||
| 抄録 | ||||||||||||||||||||||||
| 内容記述タイプ | Abstract | |||||||||||||||||||||||
| 内容記述 | Purpose To apply a generative adversarial network (GAN)-based super-resolution (SR) for accelerating magnetic resonance imaging (MRI) and evaluate the radiomic feature reproducibility of GAN-reconstructed T2*-weighted images of patients with cerebral microbleeds (CMBs). Materials and methods This prospective study included 150 patients (71 males, 79 females; median age: 66 years) with suspected or diagnosed CMBs from July to September 2021. T2*-weighted images were acquired using MRI scanners from three different vendors with a 40 % reduction in scan time from our routine clinical practice. The high-resolution images were reconstructed using the GAN-based SR model, trained on 75 cases. Two types of regions of interest (ROIs) ?those containing CMBs and those without CMBs (control)?were defined, and 91 radiomic features were extracted from the GAN-reconstructed and reference standard images. Reproducibility of radiomic features was evaluated using intraclass correlation coefficients (ICCs). Results The median ICCs for the ROIs including CMBs and for the control ROIs were 0.90 (interquartile range [IQR]: 0.84?0.95) and 0.95 (IQR: 0.90?0.97), respectively. First order and gray level co-occurrence matrix features showed higher ICC values compared to other radiomic feature categories. Spearman rank correlation coefficients indicated consistently strong correlations between GAN-reconstructed and reference standard images across all feature categories. Bland?Altman plots revealed minimal bias and narrow limits of agreement between the GAN-reconstructed and reference standard images, further supporting high reproducibility. Conclusion The GAN-based SR method demonstrated high reproducibility for radiomic features in T2*-weighted images, suggesting that MRI scan time can be significantly reduced without substantially altering radiomic feature distributions. These findings also highlight the potential of radiomic feature reproducibility to serve as a supplementary metric for evaluating the technical consistency of AI-reconstructed images. |
|||||||||||||||||||||||
| 書誌情報 |
European Journal of Radiology Artificial Intelligence 巻 3, p. 100031, 発行日 2025-07 |
|||||||||||||||||||||||
| 出版者 | ||||||||||||||||||||||||
| 出版者 | Elsevier | |||||||||||||||||||||||
| ISSN | ||||||||||||||||||||||||
| 収録物識別子タイプ | ISSN | |||||||||||||||||||||||
| 収録物識別子 | 3050-5771 | |||||||||||||||||||||||
| DOI | ||||||||||||||||||||||||
| 識別子タイプ | DOI | |||||||||||||||||||||||
| 関連識別子 | 10.1016/j.ejrai.2025.100031 | |||||||||||||||||||||||
