notesum.ai

Published at November 26

Uncertainty quantification for White Matter Hyperintensity segmentation detects silent failures and improves automated Fazekas quantification

eess.IV

cs.CV

cs.LG

I.4.10; I.4.6; I.2.10; I.2.6; J.3; G.3

Released Date: November 26, 2024

Authors: Ben Philps¹, Maria del C. Valdes Hernandez², Chen Qin³, Una Clancy², Eleni Sakka², Susana Munoz Maniega², Mark E. Bastin², Angela C. C. Jochems, Joanna M. Wardlaw², Miguel O. Bernabeu⁴, Alzheimers Disease Neuroimaging Initiative

Aff.: ¹School of Informatics, University of Edinburgh; ²Centre for Clinical Brain Sciences, University of Edinburgh; ³Department of Electrical and Electronic Engineering & I-X, Imperial College London; ⁴Centre for Medical Informatics, Usher Institute, University of Edinburgh

Arxiv: http://arxiv.org/abs/2411.17571v1

	Dice	Top Dice	AVD%	Top AVD%
SEnt	0.67 $\pm$ 0.01	—	50.2 $\pm$ 6.7	—
MC-Drop	0.68 $\pm$ 0.00	0.70 $\pm$ 0.00	46.5 $\pm$ 4.8	23.9 $\pm$ 4.2
Ens	0.69 $\pm$ 0.00	0.69 $\pm$ 0.00	$\bm{42.7}$ $\pm$ 1.7	19.2 $\pm$ 2.7
Evid	0.68 $\pm$ 0.01	—	49.2 $\pm$ 8.0	—
Ind	0.69 $\pm$ 0.01	0.69 $\pm$ 0.01	50.1 $\pm$ 7.9	48.1 $\pm$ 7.8
P-Unet	0.68 $\pm$ 0.01	0.69 $\pm$ 0.01	55.4 $\pm$ 13.0	39.2 $\pm$ 8.8
SSN	0.68 $\pm$ 0.01	0.71 $\pm$ 0.01	55.0 $\pm$ 11.5	17.3 $\pm$ 5.5
SSN-Ens	$\bm{0.70}$ $\pm$ 0.00	$\bm{0.72}$ $\pm$ 0.00	43.8 $\pm$ 3.1	$\bm{12.3}$ $\pm$ 1.5