Deep learning segmentation with uncertainty quantification for spinal tuberculosis on fat-suppressed T2-weighted MRI.
Source check: Source checked. Not independent scientific confirmation.
Finding: In a 50-case external test set, the model achieved Dice 0.858 and pixel-level AUC 0.912; predictive uncertainty correlated with pixel-level error (ρ=0.74). In a nine-physician crossover, uncertainty overlays increased trust ratings and reduced review time. Retrospective 210-patient, three-center study: internal five-fold cross-validation and a separate 50-case external test set, plus a nine-physician within-subject crossover. Clinical decision-support measures do not establish better treatment or patient outcomes.
Technology: Improved nnU-Net combines boundary-aware loss with Monte Carlo dropout to segment spinal tuberculosis lesions and estimate pixel-level uncertainty on fat-suppressed T2 MRI.
Publication: Peer-reviewed paper · Test context: retrospective or benchmark. Catalog categories, not an evidence grade.
- AI contribution
- Improved nnU-Net combines boundary-aware loss with Monte Carlo dropout to segment spinal tuberculosis lesions and estimate pixel-level uncertainty on fat-suppressed T2 MRI.
- Test and result
- In a 50-case external test set, the model achieved Dice 0.858 and pixel-level AUC 0.912; predictive uncertainty correlated with pixel-level error (ρ=0.74). In a nine-physician crossover, uncertainty overlays increased trust ratings and reduced review time.
- Limitation / still unknown
- Retrospective 210-patient, three-center study: internal five-fold cross-validation and a separate 50-case external test set, plus a nine-physician within-subject crossover. Clinical decision-support measures do not establish better treatment or patient outcomes.
Evidence sources and annotation context
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
Evidence sources: https://www.frontiersin.org/journals/physiology/articles/10.3389/fphys.2026.1885886/full
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: not linked in this catalog; this does not mean no paper exists.
Review date: 22 Nov 2026. Review scope: The publisher full text reports an uncertainty-guided nnU-Net segmentation of spinal tuberculosis lesions on FS-T2 MRI in 210 patients from three centers. On a 50-case external test set, Dice was 0.858 and pixel-level AUC 0.912; uncertainty tracked segmentation error (Spearman ρ=0.74), and a nine-physician within-subject study found higher trust scores and shorter review time with uncertainty overlays. The first located result-bearing version is the publisher article dated 2026-07-29; exact-title/DOI searches did not locate an earlier matching repository version, without asserting global priority. The retrospective test cohort and small reader study support a bounded segmentation/decision-support result, not improved patient outcomes..
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 3/5 · Validation 2/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.