Texture analysis on MR images helps predicting non-response to NAC in breast cancer

BMC Cancer

Table 5 Performance of the individual parameters measured from ROC curves (based on the Youden index for determining cut-offs)

	Se (%)	Sp (%)	AUC	Cut-offs
Energy^a	64	79	0.702	41
Entropy^a	64	79	0.696	182
Contrast	30	95	0.576	17
Homogeneity^a	58	84	0.701	144
Correlation	62	16	0.512	42
Inv. Diff. Moment^a	60	84	0.711	152
Sum average	28	90	0.527	103
Sum variance	78	42	0.583	104
Difference variance^a	60	79	0.649	86
SRE	80	42	0.569	0.004
LRE	86	37	0.569	301
GLN	74	42	0.555	621
RLN	38	90	0.579	75
RP^a	42	90	0.640	0.740
LGRE	42	90	0.630	0.800
HGRE^a	42	90	0.644	2.40
SRLGE	70	53	0.582	0.003
SRHGE	16	100	0.510	0.007
LRLGE	80	37	0.536	233
LRHGE	72	58	0.620	781
Amplitude	67	58	0.567	69.1
Wash-out	27	95	0.594	0.09
Wash-in^a	86	47	0.685	0.50
Mass^b	63	46	0.546	_
non Mass^b	63	46	0.546	_
Ki67 > 14 %^b	42	82	0.621	_
HER2 +^b	79	44	0.615	_
HR-/HER2 +^b	100	20	0.600	_

An overall better performance of GLCM compared to RLM parameters, as well as a better performance of texture and kinetic parameters compared to BI-RADS and biological parameters was observed
^aParameters performing significantly better than a random classifier (p^{(AUC > 0.5)} < 0.05)
^bCategorial variables without cut-offs

ISSN: 1471-2407