Temel Kavramlar
Expert Disagreement-Guided Uncertainty Estimation (EDUE) improves model calibration and segmentation performance in medical image analysis.
İstatistikler
Our method achieves 55% and 23% improvement in correlation on average with expert disagreements at the image and pixel levels, respectively.
EDUE has the lowest NLL value of 0.163, indicating less overconfidence compared to DE and LE models.
Alıntılar
"Uncertainty estimation methods provide potential solutions for evaluating prediction reliability."
"Models need to convey trustworthy predictive uncertainty for clinical adoption."