Computed tomography-based radiomics diagnostic approach for differential diagnosis between early- and late-stage pancreatic ductal adenocarcinoma

Table 3 The differential diagnostic ability of the selected radiomics features

Features	AUC	SEN	SPE	ACC
a_Compactness2	0.754	67.5%	77.42%	71.83%
a_GLCMEntropy_AllDirection_offset7_SD	0.698	82.5%	58.06%	71.83%
p_Correlation_angle90_offset7	0.683	60.0%	70.97%	64.79%
p_HighGreyLevelRunEmphasis_AllDirection_offset4_SD	0.661	77.5%	51.61%	66.20%
a_GLCMEntropy_AllDirection_offset4_SD	0.665	87.5%	54.84%	73.24%
a_Correlation_angle45_offset7	0.639	50.0%	77.42%	61.97%
p_Compactness2	0.66	80.0%	54.84%	69.01%
a_InverseDifferenceMoment_AllDirection_offset4_SD	0.638	75.0%	64.52%	70.42%
a_Correlation_angle135_offset4	0.677	75.0%	58.06%	67.61%

AUC: Area under the receiver operating characteristic curve; SEN: Sensitivity; SPE: Specificity; ACC: Accuracy; GLCM: Grey level co-occurrence matrix.