Low-field MRI-derived synthetic CT is tested in liver treatment replanning
Source check: Source checked. Not independent scientific confirmation.
Finding: In 37 liver cases, U-Net-generated synthetic CT images from low-field MRI were evaluated in retrospective treatment replanning. Both loss-function variants met the reported planning objectives in this source cohort. The study does not establish clinical equivalence, independent deployment safety or patient benefit.
Technology: U-Net with MSE and VGG16-based perceptual loss
Publication: Peer-reviewed paper · Test context: retrospective or benchmark. Catalog categories, not an evidence grade.
- AI contribution
- Learned MRI-to-CT synthesis supplied images for a defined dosimetric reference comparison; conventional treatment-planning software computed the plans.
- Test and result
- The 37 patients supplied 2,561 liver-region image slices. Ten-fold evaluation grouped patients, leaving 3–4 people out per fold. Synthetic images generated for held-out cases were compared with each patient’s original CT in target/organ-at-risk planning. Perceptual and MSE variants had mean absolute image errors of 35.64 and 35.67 HU, and all synthetic-CT plans met the source planning objectives.
- Limitation / still unknown
- The same small single-system cohort informed preliminary training choices, so this is not an external independent validation. Missing fine structures such as ribs and registration between separately acquired MRI/CT remain limitations. Reported nonsignificant comparisons are not equivalence tests, and visual image quality is not delivered-dose safety. Manual target contours were transferred from original CT; an entirely CT-free clinical workflow was not tested. No actual treatment-outcome comparison or demonstrated clinical benefit is reported. The August 2021 online version is represented; its issue date is 2022.
Evidence sources and annotation context
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
Evidence sources: https://pmc.ncbi.nlm.nih.gov/articles/PMC8741640/
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review date: 21 Nov 2026. Review scope: Original computational training/partition Methods, represented applied Results and relevant limitations actually read; no source code executed..
No follow-up recorded; this is not a claim that no follow-up exists. Follow-up review scope and date are not recorded. No source-supported relationships recorded.
Editorial assessment: Impact 1/5 · Validation 2/5 · Novelty 2/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.