Deep convolutional neural networks for automated segmentation of brain metastases trained on clinical data

Radiation Oncology

Table 2 Performance of the algorithms by network type and type of ensemble building. SUM: summation, MV: majority voting

DCNN Type	Ensemble	Sensitivity	Precision	F1-Score	Sensitivity Small BM	AFPR	Mean DSC	DSC
DCNN Type	Method	Sensitivity	Precision	F1-Score	Sensitivity Small BM	AFPR	Mean DSC	per Lesion
moU-Net	SUM	0.71	0.89	0.79	0.51	0.18	0.71	0.74
cU-Net	SUM	0.71	0.94	0.81	0.51	0.1	0.7	0.73
sU-Net	SUM	0.53	0.85	0.65	0.68	0.2	0.27	0.61
NetSUM		0.82	0.83	0.82	0.7	0.35	0.7	0.74
moU-Net	MV	0.65	0.96	0.78	0.43	0.05	0.71	0.73
cU-Net	MV	0.63	1	0.77	0.4	0	0.69	0.73
sU-Net	MV	0.43	0.95	0.59	0.62	0.05	0.21	0.52
NetMV		0.77	0.96	0.85	0.64	0.08	0.71	0.71

DSC dice similarity coefficient, AFPR average false positive rate, F1-score combines sensitivity and specificity into a single metric by calculation of their harmonic mean in order to find the most balanced model

ISSN: 1748-717X