Table 2 Comparison of the classified water accuracy for the volumes at time steps T1, T2 and T3 computed between the SIRT-FBP-MD-D-DIFF and target volumes used for network training and validation.

Network	Dataset	Dice			Sensitivity
Network	Dataset	T1 (%)	T2 (%)	T3 (%)	T1 (%)	T2 (%)	T3 (%)
Standard	PEFC_1, training	80	87	85	91	92	94
Standard	PEFC_1, validation	76	82	80	93	91	92
Quasi-4D	PEFC_1, training	83	89	86	93	95	92
Quasi-4D	PEFC_1, validation	77	84	81	94	93	88
Ensemble	PEFC_1, training	82	88	87	88	87	89
Ensemble	PEFC_1, validation	79	82	81	89	83	83

The target volumes were computed with the rSIRT-PWC-DIFF algorithm¹⁷. The results are shown separately for each MS-D network design, further split between the training and validation halves of the PEFC_1 fuel cell dataset. The classified water accuracy was measured through dice coefficient and sensitivity through the volume.

Quick links

Search