Diagnostic utility of a conventional MRI-based analysis and texture analysis for discriminating between ovarian thecoma-fibroma groups and ovarian granulosa cell tumors

Table 5 Representative texture features selected using the LASSO algorithm

Feature name code	Value*, mean ± SD		Univariate analysis (p)	LASSO estimate	ICC
Feature name code	OTFG	OGCT	Univariate analysis (p)	LASSO estimate	ICC
Diff.variance	-0.20 ± 0.48	0.64 ± 1.17	0.012	0.163	0.757
10th percentile	-0.34 ± 0.65	0.38 ± 0.52	< 0.001	0.104	0.988
LAHGLE	0.18 ± 0.80	1.28 ± 2.40	0.015	0.056	0.792
Strength	-0.01 ± 0.87	0.72 ± 1.14	0.007	0.040	0.578
Zone variance	0.43 ± 1.36	1.39 ± 4.38	0.993	0.027	0.944
Joint energy	0.13 ± 1.01	0.34 ± 1.57	0.834	0.002	0.950
Busyness	1.46 ± 3.21	-0.29 ± 0.67	< 0.001	-0.007	0.806
Total energy	0.30 ± 1.04	0.35 ± 1.61	0.617	-0.018	0.952
SDLGLE	2.31 ± 6.20	0.03 ± 0.84	0.003	-0.025	0.991
MCC	-0.43 ± 1.74	-0.70 ± 1.58	0.248	-0.088	0.779

All 93 texture features (TFs) were included in the LASSO regression analysis. Features with a coefficient other than 0 are shown in this table. Regarding all variables, numeric values were standardized by RobustScaler before the statistical analysis (* standardized values are shown). ICC intraclass correlation coefficient, LASSO least absolute shrinkage and selection operator, SD standard deviation, OTFG ovarian thecoma-fibroma group, OGCT ovarian granulosa cell tumor. Feature name codes are as follows: Diff.variance difference variance, LAHGLE large area high gray-level emphasis, MCC maximal correlation coefficient, SDLGLE small dependence low gray-level emphasis

ISSN: 1757-2215