Research Article

Distinguishing Lymphomatous and Cancerous Lymph Nodes in 18F-Fluorodeoxyglucose Positron Emission Tomography/Computed Tomography by Radiomics Analysis

Table 5

The differentiating ability of each predictive variable in cross-validation (training set and test set).

In training set
Predictive variablesAUCCutoff pointSensitivity (%)Specificity (%)Accuracy (%)

LLN versus CLNSUVmax0.7410.4084.5261.4572.46<0.01
PREct0.920.3595.2483.1389.22<0.01
PREpet0.900.4282.1485.5483.83<0.01
PREcombination0.960.3992.8693.9892.81<0.01
HLLN versus CLNSUVmax0.727.1679.0758.5465.62<0.01
PREct0.950.3388.3798.7895.24<0.01
PREpet0.900.6379.0785.3785.67<0.01
PREcombination0.980.6297.6798.7896.83<0.01
NHLLN versus CLNSUVmax0.7110.4085.5453.0169.28<0.01
PREct0.900.4195.1883.1388.55<0.01
PREpet0.860.4784.3478.3180.72<0.01
PREcombination0.960.2887.9592.7789.16<0.01
HLLN versus NHLLNSUVmax0.5211.2046.9963.4155.15>0.05
PREct0.790.3681.9370.7374.19<0.01
PREpet0.790.4181.9365.8579.03<0.01
PREcombination0.860.5192.7768.2983.87<0.01

In test set
LLN versus CLNSUVmax0.6710.4080.5637.1459.15<0.01
PREct0.960.3588.8991.4390.14<0.01
PREpet0.870.4277.7871.4374.65<0.01
PREcombination0.950.3991.6794.2992.96<0.01
HLLN versus CLNSUVmax0.657.1672.2252.7859.26<0.05
PREct0.830.3366.6797.2287.04<0.01
PREpet0.760.6361.1175.0070.37<0.01
PREcombination0.850.6277.7897.2290.74<0.01
NHLLN versus CLNSUVmax0.7010.4082.8647.2264.79<0.01
PREct0.840.4197.1475.0085.92<0.01
PREpet0.880.4780.0075.0077.46<0.01
PREcombination0.900.2877.1488.8983.10<0.01
HLLN versus NHLLNSUVmax0.5711.2047.2276.4756.60>0.05
PREct0.740.3672.2252.9466.04<0.01
PREpet0.720.4180.5658.8273.58<0.01
PREcombination0.780.5188.8952.9477.36<0.01

The cutoff point in the training set is consistent with that in the verification set. Abbreviations: AUC, area under the ROC curve; ROC, receiver operating characteristics; SUV, standardized uptake value; LLN, lymphomatous-lymph nodes; CLN, cancerous-lymph nodes; HLLN, LLN caused by Hodgkin’s lymphoma; NHLLN, LLN caused by non-Hodgkin lymphoma; PREct, CT-predictive variables; PREpet, PET-predictive variables, PREcombination, the combination of PET and CT predictive variables.