MDCT-Based Radiomics Features for the Differentiation of Serous Borderline Ovarian Tumors and Serous Malignant Ovarian Tumors

Figures & data

Figure 1 Workflow of radiomics analysis.

Abbreviations: ICC, intraclass correlation coefficient; LASSO, least absolute shrinkage, and selection operator; ROC, receiver operating characteristic.

Table 1 The Patient Characteristics of SBOTs and SMOTs

	Training Set (n =127)			Validation Set (n = 55)
	SBOTs (n =56)	SMOTs (n = 71)	p	SBOTs (n =24)	SMOTs (n = 31)	p
Age (years, mean ±std)	41.2 ± 11.2	52.5 ± 9.0	<0.001	40.2 ± 15.2	54.5 ± 10.0	<0.001
Maximum diameter (cm, mean ± std)	6.0 ± 3.4	8.1± 4.4	0.112	6.7 ± 4.4	9.1 ±3.7	0.098

Abbreviations: SBOTs, serous borderline ovarian tumors; SMOTs, serous malignant ovarian tumors.

Table 2 The Most Predictive Features Between SBOTs and SMOTs Selected by LASSO Regression

Phase	Feature Name	Feature Class	Regression Coefficient
AP	LargeDependenceEmphasis	GLDM	0.329
	ZoneEntropy	GLSZM	0.325
	RunLengthNonUniformityNormalized	GLRLM	−0.273
	GrayLevelNonUniformityNormalized	GLSZM	0.228
	IMC2	GLCM	0.215
	MCC	GLCM	0.174
	LowGrayLevelRunEmphasis	GLRLM	0.159
	MaximumProbability	GLCM	−0.156
	RootMeanSquared	FIRSTORDER	0.132
VP	IDN	GLCM	0.692
	RunLengthNonUniformityNormalized	GLRLM	−0.584
	IMC2	GLCM	0.563
	ClusterShade	GLCM	0.476
	InverseVariance	GLCM	0.347
	Contrast	NGTDM	−0.322
	LowGrayLevelZoneEmphasis	GLSZM	0.321
	Median	FIRSTORDER	−0.318
	GrayLevelNonUniformityNormalized	GLSZM	0.305
EP	RunLengthNonUniformityNormalized	GLRLM	−0.598
	Median	FIRSTORDER	−0.578
	IMC2	GLCM	−0.570
	MCC	GLCM	0.569
	GrayLevelNonUniformityNormalized	GLSZM	0.549
	GrayLevelNonUniformityNormalized	GLRLM	−0.457
	ClusterShade	GLCM	0.447
	LowGrayLevelZoneEmphasis	GLSZM	0.430
	IDMN	GLCM	0.428

Abbreviations: SBOTs, serous borderline ovarian tumors; SMOTs, serous malignant ovarian tumors; AP, arterial phase; VP, venous phase; EP, equilibrium phase.

Figure 2 The ROC curves and AUC values of the most predictive features between SBOTs and SMOTs with LASSO regression in AP (A), VP (B), and EP (C). The numbers in each figure mean the radiomics parameters in.Table 2

Abbreviations: ROC, receiver operating characteristic; AUC, area under curve; SBOTs, serous borderline ovarian tumors; SMOTs, serous malignant ovarian tumors; LASSO, least absolute shrinkage, and selection operator; AP, arterial phase; VP, venous phase; EP, equilibrium phase.

Figure 3 The heat maps of radiomics parameter distribution of the AP, VP and EP between SBOTs and SMOTs. Difference in colors means different value of radiomics parameter.

Abbreviations: AP, arterial phase; VP, venous phase; EP, equilibrium phase; SBOTs, serous borderline ovarian tumors; SMOTs, serous malignant ovarian tumors.

Table 3 The Predictive Performance of AP, VP, and EP Model

	AUC	Accuracy	Sensitivity	Specificity
AP model	0.80	0.75	0.74	0.75
VP model	0.86	0.78	0.80	0.75
EP model	0.73	0.69	0.71	0.67

Abbreviations: AUC, area under curve; AP, arterial phase; VP, venous phase; EP, equilibrium phase.

Table 4 The Results of Multiple Comparisons of the AUCs by the Delong Test

	Z Statistic	p
AP-VP	1.283	0.199
AP-EP	1.127	0.260
VP-EP	0.262	0.793

Abbreviations: AUC, area under curve; AP, arterial phase; VP, venous phase; EP, equilibrium phase.

Figure 4 The ROC curves and AUC values of the AP, VP, and EP to differentiate SBOTs and SMOTs.

Abbreviations: ROC, receiver operating characteristic; AUC, area under curve; AP, arterial phase; VP, venous phase; EP, equilibrium phase; SBOTs, serous borderline ovarian tumors; SMOTs, serous malignant ovarian tumors.