Deep learning and radiomics framework for PSMA-RADS classification of prostate cancer on PSMA PET

EJNMMI Research

Table 2 Performance on PSMA-RADS classification

Inputs	AUROC	Accuracy	Precision	Recall	F1 score
Validation set: Lesion-level performance
Manual	0.95 (0.95, 0.95)	0.71 (0.68, 0.74)	0.71	0.71	0.71
Predicted RF	0.95 (0.95, 0.95)	0.70 (0.67, 0.73)	0.71	0.70	0.70
Predicted TT	0.93 (0.93, 0.93)	0.68 (0.64, 0.71)	0.67	0.68	0.67
Predicted RF + TT	0.93 (0.93, 0.93)	0.67 (0.64, 0.70)	0.67	0.67	0.67
Test set: Lesion-level performance
Manual	0.91 (0.91, 0.91)	0.61 (0.58, 0.65)	0.62	0.61	0.61
Predicted RF	0.90 (0.90, 0.90)	0.56 (0.53, 0.60)	0.58	0.56	0.57
Predicted TT	0.88 (0.88, 0.88)	0.55 (0.52, 0.59)	0.56	0.55	0.55
Predicted RF + TT	0.87 (0.87, 0.88)	0.52 (0.48, 0.56)	0.53	0.52	0.52
Test set: Patient-level performance
Manual	0.91 (0.90, 0.91)	0.77 (0.66, 0.89)	0.79	0.77	0.76
Predicted RF	0.87 (0.87, 0.88)	0.68 (0.55, 0.80)	0.69	0.68	0.68
Predicted TT	0.92 (0.92, 0.92)	0.81 (0.71, 0.92)	0.85	0.81	0.82
Predicted RF + TT	0.90 (0.90, 0.90)	0.77 (0.66, 0.89)	0.78	0.77	0.77

Data in parenthesis correspond to 95% confidence intervals. Manual refers to using the radiomic features extracted from manual segmentations and the manually annotated tissue types as inputs. Predicted refers to using the automatically extracted radiomic features and the automatically predicted tissue types as inputs. AUROC = area under the receiver operating characteristic. RF = radiomic features. TT = tissue types