Table 1 Model performance and concordance with pathologist/surgeon annotations for the gross measurements, tissue orientation/mapping, completeness, and margin assessment tasks.

Task	Evaluation metric	Subtask	Estimate	2.5% CI	97.5% CI
Gross Measurements	3D Point Cloud: MAD^a (cm)	L	0.36	0.21	0.53
		W	0.23	0.12	0.52
		H	0.29	0.19	0.50
		Overall	0.29	0.2	0.47
	NeRF: MAD (cm)	L	0.21	0.11	0.53
		W	0.19	0.17	0.22
		H	0.27	0.17	0.36
		Overall	0.22	0.18	0.33
	3D Point Cloud: MPC^b (%)	L	12.4	7.9	17.7
		W	13.0	8.3	32.0
		H	31.7	20.0	58.0
		Overall	19.0	12.6	33.2
	NeRF: MPC (%)	L	11.3	4.5	14.3
		W	11.3	8.3	16.9
		H	34	6.5	58.3
		Overall	18.9	8.2	28.5
	3D Point Cloud: Correlation	L	0.848	0.544	0.973
		W	0.833	0.467	0.971
		H	0.812	0.456	0.955
		Overall	0.831	0.602	0.937
	NeRF: Correlation	L	0.946	0.814	0.988
		W	0.875	0.576	0.976
		H	0.893	0.646	0.984
		Overall	0.905	0.782	0.96
Tissue orientation	MAD (°)		4.883	4.04	5.451
	Proportion Correct Orientation	≤45°	94.7%	92.3%	96.6%
		≤15°	86.0%	82.6%	88.9%
		≤5°	51.3%	46.7%	55.7%
GNN: Tissue completeness	AUC		0.839	0.825	0.855
	Macro-AUC		0.851	0.839	0.863
GNN: Margin assessment	AUC	Original Slides	0.967	0.960	0.979
		Follicle Removal	0.965	0.959	0.975
	Macro-AUC	Original Slides	0.962	0.954	0.970
		Follicle Removal	0.957	0.949	0.964
CNN: Tissue completeness	AUC		0.923	0.917	0.930
	Macro-AUC		0.916	0.907	0.925
CNN: Margin assessment	AUC		0.900	0.887	0.912
	Macro-AUC		0.877	0.859	0.893
Tumor Mapping	Accuracy		0.992	0.915	0.999

Macro-AUC represents reporting of AUC statistic on slide level and averaging across slides, giving each slide equal weight, while normal AUC statistic is calculated for subimages across all slides. 95% confidence intervals were acquired using 1000-sample non-parametric bootstrap, where bootstrapping was done on the WSI level to account for variation in concordance across the cases.
^aMedian Absolute Deviation.
^bMedian Proportion Change.

Quick links

Search