Insights into Imaging

Table 3 Impact of various factors on segmentation variability with 2 methods (pair-wise comparison and consensus comparison (STAPLE reference))

From: Challenge of prostate MRI segmentation on T2-weighted images: inter-observer variability and impact of prostate morphology

		Pair-wise comparison			STAPLE reference
Factor	Metric	ρ^a	CI 95%	p-value	ρ^a	CI 95%	p-value
Volume^a	DSC	0.84	[0.72; 0.91]	< 0.001	0.86	[0.76; 0.93]	< 0.001
	HD (voxels)	0.28	[− 0.04; 0.54]	0.5	0.20	[− 0.12; 0.48]	1.0
	AHD (voxels)	− 0.42	[− 0.65; − 0.13]	0.06	− 0.63	[− 0.78; − 0.39]	< 0.001
Squared TZ to PZ contrast^a	DSC	0.50	[0.23; 0.71]	0.01	0.45	\|0.17; 0.67]	0.03
	HD (voxels)	− 0.03	[− 0.34; 0.29]	1.0	− 0.04	[− 0.34; 0.28]	1.0
	AHD (voxels)	− 0.49	[− 0.7; − 0.22]	0.01	− 0.45	[− 0.67; − 0.17]	0.03
Presence of a PI-RADS ≥ 3 lesion^b	DSC			0.53			0.71
	HD (voxels)			0.3			0.47
	AHD (voxels)			1.0			1.0
Large median lobe^b	DSC			0.08			0.11
	HD (voxels)			0.61			1.0
	AHD (voxels)			1.0			1.0

^aTest: spearman correlation
^bTest: Mann-u-Whitney

Back to article page