Disentangling weak spatial priors from generic local context in radiographic abnormality detection: a controlled mechanism audit
Biomedical Physics & Engineering Express · 2026-12-01 · 12 (6) · 065018
DOI: 10.1088/2057-1976/aebbbf ↗Abstract
Abstract Background. Improvements from spatially guided crops may reflect generic local context, optimization or operating-point selection rather than informative spatial guidance. Purpose. To assess whether a coarse foreground guidance map improves radiographic abnormality discrimination beyond non-prior local controls. Methods. We audited original frozen DenseNet121 pipelines and conducted a revision-stage matched experiment using five seeds, patient-disjoint development partitions, and prior-guided, centered and random-position crops. Crops retained 90% of image width and height. The primary revised comparison used a common eight-epoch local budget and study ROC-AUC. Validation-selected checkpoints and fixed 50:50 global–local fusions were secondary. Evaluation used 1199 studies from 783 patients, with paired patient-cluster bootstrap intervals. This cohort had already been evaluated in the original study. Results. At the common final epoch, mean ROC-AUC was 0.892 269 for prior-guided, 0.891 573 for centered and 0.892 221 for random-position crops. Mean paired AUC differences were +0.000 696 (seed SD 0.002 757) versus centered and +0.000 048 (SD 0.002 693) versus random. All ten primary per-seed 95% intervals included zero. Fallback exclusion and alternative aggregation yielded small descriptive differences. Original fusion pipelines had similar AUC but different operating profiles at their independently frozen thresholds. Conclusion. Within MURA and this conservative DenseNet121 crop procedure, spatial guidance showed no consistent ranking advantage over the tested controls. The findings do not establish equivalence or exclude benefits from stronger or different priors. Matching training exposure and reporting discrimination, calibration and operating-point behavior separately strengthen the audit without establishing clinical trustworthiness.
EUVIMED