Predicting response to CCRT for esophageal squamous carcinoma by a radiomics-clinical SHAP model

BMC Medical Imaging

Table 6 Model performance comparison

Model	AUC (95% CI)	Accuracy	Precision	Recall	F1-Score
training set:
Clinical	0.89 (0.81–0.95)	0.81	0.80	0.96	0.88
TNM	0.46(0.25–0.68)	0.73	0.73	0.96	0.83
CTV	0.89 (0.80–0.96)	0.83	0.81	0.99	0.88
GTV	0.83 (0.72–0.93)	0.76	0.73	1.0	0.85
CTV-Clinical	0.67 (0.56–0.78)	0.76	0.76	0.94	0.84
GTV-Clinical	0.99 (0.98–0.99)	0.96	0.94	1.0	0.97
testing set:
Clinical	0.80 (0.60–0.96)	0.75	0.72	1.0	0.84
TNM	0.64 (0.51–0.77)	0.54	0.62	0.71	0.70
CTV	0.70(0.43–0.94)	0.63	0.67	0.88	0.76
GTV	0.77 (0.53–0.96)	0.75	0.75	0.94	0.83
CTV-Clinical	0.56 (0.4–0.75)	0.67	0.70	0.86	0.78
GTV-Clinical	0.82 (0.66–0.99)	0.71	0.74	0.88	0.80
validation set:
Clinical	0.82 (0.57–1.0)	0.55	0.44	1.0	6.15
TNM	0.80 (0.50–1.0)	0.64	0.50	1.0	0.67
CTV	0.68 (0.25–1.0)	0.45	0.4	1.0	0.57
GTV	0.91 (0.67–1.0)	0.73	0.57	1.0	0.73
CTV-Clinical	0.57 (0.50–0.75)	0.54	0.44	1.0	0.62
GTV-Clinical	0.97 (0.84–1.0)	0.72	0.57	1.0	0.73

ISSN: 1471-2342