Clinico-biological-radiomics (CBR) based machine learning for improving the diagnostic accuracy of FDG-PET false-positive lymph nodes in lung cancer

Table 3 Performance of models for predicting hypermetabolic mediastinal–hilar LNs status in lung cancer

Models	AUC (95% CI)	SEN	SPE	ACC	PPV	NPV	FPR	FNR
Training set
LN SUVmax	0.83 (0.77–0.89)	79.81	71.79	76.37	79.05	72.73	28.21	20.19
LN enlarged	0.71 (0.65–0.78)	71.15	71.79	71.43	77.08	65.12	28.21	28.85
LN_PET/CT	0.83 (0.77–0.89)	67.31	85.90	75.27	86.42	66.34	14.10	32.69
CBI model	0.74 (0.66–0.81)	75.00	43.59	61.54	63.93	56.67	56.41	25.00
Rad model	0.76 (0.68–0.83)	85.58	60.26	74.73	74.17	75.81	39.74	14.42
CBR model	0.90 (0.86–0.95)	81.73	87.18	84.07	89.47	78.16	12.82	18.27
Test set
LN SUVmax	0.76 (0.65–0.87)	85.11	61.29	75.64	76.92	73.08	38.71	14.89
LN enlarged	0.71 (0.61–0.80)	57.45	83.87	67.95	84.38	56.52	16.13	42.55
LN_PET/CT	0.76 (0.66–0.87)	55.32	93.55	70.51	92.86	58.00	6.45	44.68
CBI model	0.73 (0.62–0.85)	76.60	58.06	69.23	73.47	62.07	41.94	23.40
Rad model	0.83 (0.73–0.93)	89.36	70.97	82.05	82.35	81.48	29.03	10.64
CBR model	0.89 (0.82–0.96)	74.47	93.55	82.05	94.59	70.73	6.45	25.53

AUC area under the receiver operating curve, CI confidence interval, SEN sensitivity, SPE specificity, ACC accuracy, PPV positive predictive value, NPV negative predictive value, FPR false positive rate, FNR false negative rate, LN lymph node, SUV standardized uptake value, PET/CT positron emission tomography/computed tomography, CBI clinico-biological-image, Rad radiomics, CBR clinico-biological-radiomics

ISSN: 2047-783X