Scalable robust graph and feature extraction for arbitrary vessel networks in large volumetric datasets

BMC Bioinformatics

Table 2 Results of the robustness test [41] using 36 rotations around the z-axis for each respective dataset

	Property	\(\mathcal {G}_{length}\)	\(\mathcal {G}_{roundMean}\)	\(\mathcal {G}_{straight}\)	\(\mathcal {N}_{FNR}\)	\(\mathcal {N}_{FPR}\)
Dataset	Refinement
Lymphatic 1	Iterative Refinement	0.781	0.780	0.823	0.0345	0.0355
Lymphatic 1	No Refinement	0.450	0.500	0.578	0.0515	0.0505
Lymphatic 2	Iterative Refinement	0.736	0.775	0.785	0.0328	0.0341
Lymphatic 2	No Refinement	0.486	0.536	0.614	0.0550	0.0539
Lymphatic 3	Iterative Refinement	0.725	0.760	0.762	0.0431	0.0386
Lymphatic 3	No Refinement	0.487	0.516	0.612	0.0653	0.0632
Synthetic 1	Iterative Refinement	0.862	0.600	0.916	0.0445	0.0477
Synthetic 1	No Refinement	0.623	0.446	0.696	0.0408	0.0488
Synthetic 2	Iterative Refinement	0.819	0.598	0.895	0.0425	0.0461
Synthetic 2	No Refinement	0.603	0.452	0.690	0.0393	0.0479

\(\mathcal {G}_{property}\) denotes the GERoMe index for the given property. \(\mathcal {N}_{FNR}\) and \(\mathcal {N}_{FPR}\) use the same perturbation procedure, but measure the geometric error of NetMets [42] of the calculated centerlines and therefore show the aggregated maximum value. Refined graphs were extracted using a bulge size of 1.5 and iteration until reaching a fixed point. Relative score differences above 10% are highlighted as bold text

ISSN: 1471-2105