Table 2.

Model performance

ModelFeatures AUC training cohort (95% CI)CV-AUC training cohort (95% CI)AUC validation cohort (95% CI)
Clinical Ann Arbor stage
B symptoms
R/R status 
0.787 (0.692-0.883) 0.729 (0.724 - 0.734) 0.677 (0.535-0.819) 
Radiomics SUV 4.0 No. of lesions
VolSpread
TLRSUVmean 
0.719 (0.605-0.833) 0.691 (0.685-0.696) 0.721 (0.580-0.863) 
Final model R/R status
B symptoms
MTV
Spread
TLRSUVmean 
0.837 (0.744-0.930) 0.810 (0.805-0.814) 0.750 (0.627-0.872) 
P value of clinical vs final model  .00094 .0049 <.0001 
ModelFeatures AUC training cohort (95% CI)CV-AUC training cohort (95% CI)AUC validation cohort (95% CI)
Clinical Ann Arbor stage
B symptoms
R/R status 
0.787 (0.692-0.883) 0.729 (0.724 - 0.734) 0.677 (0.535-0.819) 
Radiomics SUV 4.0 No. of lesions
VolSpread
TLRSUVmean 
0.719 (0.605-0.833) 0.691 (0.685-0.696) 0.721 (0.580-0.863) 
Final model R/R status
B symptoms
MTV
Spread
TLRSUVmean 
0.837 (0.744-0.930) 0.810 (0.805-0.814) 0.750 (0.627-0.872) 
P value of clinical vs final model  .00094 .0049 <.0001 

Spread, the sum of the distances between all lesions; TLRSUVmean, tumor-to-liver ratio of the lesion SUVmean and liver SUVmean; VolSpread, the sum of differences in volume between all lesions.

All radiomics variables are log transformed.

P values represent the added value of the radiomics features to the clinical model. P value of the cross-validated (CV)-AUC represents the median P value of 2000 repeats of fivefold of cross-validation.

Close Modal

or Create an Account

Close Modal
Close Modal