Prediction of the local treatment outcome in patients with oropharyngeal squamous cell carcinoma using deep learning analysis of pretreatment FDG-PET images

Table 2 Diagnostic performance obtained by ROC curve analyses

Classifier	AUC	Sensitivity	Specificity	PPV	NPV	Accuracy
DL AlexNet
Ax	0.62 (0.45, 0.79)	0.47	0.77	0.5	0.75	0.67
Cor	0.61 (0.43, 0.77)	0.35	0.86	0.55	0.73	0.69
Ax and Cor combined	0.69 (0.54, 0.84)	0.71	0.69	0.52	0.83	0.69
DL GoogLeNet
Ax	0.72 (0.57, 0.88)	0.65	0.8	0.61	0.82	0.75
Cor	0.68 (0.52, 0.84)	0.53	0.83	0.6	0.78	0.73
Ax and Cor combined	0.8 (0.67, 0.92)	0.94	0.71	0.62	0.96	0.79
DL ResNet
Ax	0.72 (0.56, 0.88)	0.59	0.86	0.67	0.81	0.77
Cor	0.69 (0.53, 0.86)	0.47	0.91	0.73	0.78	0.77
Ax and Cor combined	0.85 (0.73, 0.96)	0.88	0.8	0.68	0.93	0.83
Conventional method
T-stage	0.62 (0.47, 0.78)	0.76	0.43	0.4	0.79	0.54
Clinical stage	0.59 (0.43, 0.74)	0.82	0.4	0.4	0.82	0.54
Multivariate clinical model	0.74 (0.61, 0.89)	0.76	0.62	0.5	0.84	0.67

Data in parentheses are 95% confidence intervals, AUC Area under curve, DL Deep learning, Ax Axial, Cor Coronal, PPV Positive predictive value, NPV Negative predictive value

ISSN: 1471-2407