What does this research mean for the field?

Deep learning-based MRI reconstruction significantly alters radiomic feature values and reduces test-retest repeatability, demonstrating that DL-reconstructed and conventional images cannot be used interchangeably for quantitative analysis. Novelty: ClaimNovelty.NOVEL_FINDING. Consensus alignment: ConsensusAlignment.NEUTRAL.

What question did this study set out to answer?

To evaluate how deep learning-based MRI reconstruction affects radiomic feature values, intraobserver reproducibility, and test–retest repeatability across multiple brain regions.

August 29, 2026Scientific ReportsOpen Access

Deep learning MRI reconstruction significantly alters radiomic features and affects their reproducibility and repeatability

View Full Paper

Ask AI

Bookmark

Authors

YJYeon Jin Jo KLKija Lee SLSang‐Kwon Lee

Discussion

Member takes

Overview

Exploratory imaging study reveals altered radiomic feature values and reduced repeatability following deep learning MRI reconstruction in canine brains, indicating caution before clinical integration.

Key Points

To evaluate how deep learning-based MRI reconstruction affects radiomic feature values, intraobserver reproducibility, and test–retest repeatability across multiple brain regions.
Acquired T1-weighted, T2-weighted, and FLAIR MRI sequences twice in nine healthy dogs (N=9).
Extracted radiomic features from six anatomical brain regions using conventional and deep learning-reconstructed scans.
Assessed feature value changes using Wilcoxon signed-rank tests, intraobserver reproducibility via intraclass correlation coefficients (ICCs), and test–retest repeatability through coefficients of variation (CVs).
Deep learning reconstruction significantly altered the majority of radiomic feature values across all sequences and anatomical regions compared to conventional reconstruction.
Images generated with deep learning reconstruction generally exhibited higher intraobserver ICCs across multiple features and regions.
Deep learning-reconstructed scans demonstrated higher CVs across sequences, reflecting lower test–retest repeatability compared to conventional images.

Cite This Study

Jo et al. (2026) studied this question.

synapsesocial.com/papers/6a9299088e5d7d1fc0c10c6c https://doi.org/10.1038/s41598-026-68549-9

View Full Paper

Ask AI

Bookmark

August 29, 2026Scientific ReportsOpen Access

Deep learning MRI reconstruction significantly alters radiomic features and affects their reproducibility and repeatability

View Full Paper

Ask AI

Bookmark

Authors

YJYeon Jin Jo KLKija Lee SLSang‐Kwon Lee

Discussion

Member takes

Overview

Key Points

To evaluate how deep learning-based MRI reconstruction affects radiomic feature values, intraobserver reproducibility, and test–retest repeatability across multiple brain regions.
Acquired T1-weighted, T2-weighted, and FLAIR MRI sequences twice in nine healthy dogs (N=9).
Extracted radiomic features from six anatomical brain regions using conventional and deep learning-reconstructed scans.
Assessed feature value changes using Wilcoxon signed-rank tests, intraobserver reproducibility via intraclass correlation coefficients (ICCs), and test–retest repeatability through coefficients of variation (CVs).
Deep learning reconstruction significantly altered the majority of radiomic feature values across all sequences and anatomical regions compared to conventional reconstruction.
Images generated with deep learning reconstruction generally exhibited higher intraobserver ICCs across multiple features and regions.
Deep learning-reconstructed scans demonstrated higher CVs across sequences, reflecting lower test–retest repeatability compared to conventional images.

Cite This Study

Jo et al. (2026) studied this question.

synapsesocial.com/papers/6a9299088e5d7d1fc0c10c6c https://doi.org/10.1038/s41598-026-68549-9

View Full Paper

Ask AI

Bookmark