Uncertainty-aware feature mapping and test-time adaptation improve brain tumor segmentation when contrast-enhanced T1 is missing

Missing contrast-enhanced T1 MRI degrades brain tumor segmentation. UAF-AIMM framework uses feature mapping, uncertainty attention, and test-time adaptation, achieving 89.2% Dice and 5.5 mm HD95 on BraTS 2021, with greatest gains in enhancing tumors.

• Retrospective evaluation on the BraTS 2021 public dataset; no external or prospective validation.
• Improvement was most pronounced in the contrast-dependent enhancing tumor subregion (exact figures not reported).
• Single public dataset; lacks external validation to confirm generalizability.

RadPigeon summaries are original and for information only. They are not clinical advice.