Inferring FDG-PET-positivity of lymph node metastases in proven lung cancer from contrast-enhanced CT using radiomics and machine learning

Gorodetski, Boris; Becker, Philipp Hendrik; Baur, Alexander Daniel Jacques; Hartenstein, Alexander; Rogasch, Julian Manuel Michael; Furth, Christian; Amthauer, Holger; Hamm, Bernd; Makowski, Marcus; Penzkofer, Tobias

doi:10.1186/s41747-022-00296-8

European Radiology Experimental

Table 2 Performance comparison of radiomics after bagging with 30 times of bootstrap iterations

From: Inferring FDG-PET-positivity of lymph node metastases in proven lung cancer from contrast-enhanced CT using radiomics and machine learning

Classifier	FSM	Accuracy	Sensitivity	Specificity	PPV	NPV	AUC
LDA	AUC	0.79 (0.77−0.81)	0.74 (0.71−0.77)	0.8 (0.78−0.84)	0.57 (0.55−0.62)	0.9 (0.89−0.91)	0.86 (0.85−0.87)
LDA	MI	0.79 (0.77−0.8)	0.73 (0.7−0.76)	0.8 (0.79−0.83)	0.58 (0.55−0.61)	0.89 (0.88−0.9)	0.86 (0.85−0.87)
LDA	MRMI	0.78 (0.76−0.8)	0.72 (0.7−0.77)	0.8 (0.75−0.83)	0.56 (0.53−0.6)	0.89 (0.88−0.9)	0.86 (0.85−0.87)
LDA	Wilcoxon	0.79 (0.76−0.81)	0.76 (0.72−0.79)	0.79 (0.76−0.82)	0.57 (0.53−0.61)	0.9 (0.89−0.91)	0.86 (0.86−0.87)
Logistic regression	AUC	0.78 (0.76−0.8)	0.73 (0.69−0.76)	0.8 (0.78−0.82)	0.57 (0.54−0.59)	0.89 (0.88−0.9)	0.87 (0.85−0.87)
Logistic regression	MI	0.78 (0.71−0.81)	0.73 (0.63−0.81)	0.8 (0.69−0.87)	0.57 (0.48−0.66)	0.89 (0.86−0.92)	0.87 (0.74−0.88)
Logistic regression	MRMI	0.78 (0.74−0.8)	0.73 (0.7−0.79)	0.79 (0.75−0.82)	0.56 (0.51−0.6)	0.89 (0.88−0.91)	0.86 (0.83−0.87)
Logistic regression	Wilcoxon	0.78 (0.77−0.81)	0.74 (0.71−0.77)	0.79 (0.78−0.83)	0.57 (0.54−0.61)	0.89 (0.89−0.91)	0.86 (0.85−0.87)
MLP	AUC	0.78 (0.73−0.81)	0.73 (0.69−0.83)	0.79 (0.71−0.85)	0.56 (0.49−0.63)	0.89 (0.88−0.92)	0.86 (0.85−0.87)
MLP	MI	0.78 (0.73−0.82)	0.71 (0.68−0.82)	0.81 (0.73−0.86)	0.57 (0.5−0.64)	0.89 (0.87−0.92)	0.86 (0.85−0.87)
MLP	MRMI	0.78 (0.74−0.82)	0.73 (0.68−0.8)	0.8 (0.73−0.84)	0.57 (0.51−0.64)	0.89 (0.87−0.91)	0.86 (0.84−0.87)
MLP	Wilcoxon	0.78 (0.71−0.8)	0.74 (0.68−0.8)	0.79 (0.68−0.84)	0.56 (0.47−0.61)	0.89 (0.88−0.91)	0.86 (0.82−0.88)
PLS	AUC	0.78 (0.75−0.8)	0.71 (0.7−0.76)	0.8 (0.75−0.83)	0.57 (0.52−0.6)	0.89 (0.88−0.9)	0.86 (0.85−0.87)
PLS	MI	0.79 (0.76−0.81)	0.71 (0.7−0.75)	0.81 (0.77−0.84)	0.57 (0.54−0.62)	0.89 (0.88−0.9)	0.86 (0.85−0.87)
PLS	MRMI	0.78 (0.75−0.8)	0.73 (0.71−0.79)	0.8 (0.75−0.83)	0.57 (0.52−0.61)	0.89 (0.88−0.91)	0.87 (0.86−0.87)
PLS	Wilcoxon	0.78 (0.74−0.79)	0.73 (0.71−0.77)	0.78 (0.75−0.82)	0.55 (0.51−0.59)	0.89 (0.88−0.91)	0.86 (0.86−0.87)
Rpart	AUC	0.74 (0.7−0.83)	0.77 (0.63−0.81)	0.73 (0.66−0.9)	0.51 (0.46−0.7)	0.9 (0.87−0.92)	0.75 (0.75−0.83)
Rpart	MI	0.77 (0.71−0.83)	0.73 (0.62−0.83)	0.78 (0.67−0.9)	0.55 (0.48−0.68)	0.89 (0.87−0.92)	0.76 (0.74−0.82)
Rpart	MRMI	0.76 (0.72−0.82)	0.73 (0.62−0.8)	0.78 (0.7−0.9)	0.54 (0.48−0.68)	0.89 (0.87−0.91)	0.76 (0.74−0.81)
Rpart	Wilcoxon	0.78 (0.69−0.81)	0.73 (0.68−0.8)	0.79 (0.66−0.83)	0.56 (0.44−0.62)	0.89 (0.88−0.92)	0.79 (0.75−0.84)
SVM	AUC	0.78 (0.76−0.8)	0.71 (0.68−0.75)	0.8 (0.78−0.84)	0.56 (0.53−0.6)	0.88 (0.88−0.9)	0.87 (0.86−0.87)
SVM	MI	0.78 (0.76−0.8)	0.71 (0.68−0.74)	0.81 (0.77−0.84)	0.57 (0.53−0.61)	0.89 (0.88−0.89)	0.86 (0.86−0.87)
SVM	MRMI	0.78 (0.75−0.8)	0.71 (0.7−0.78)	0.8 (0.76−0.83)	0.56 (0.52−0.6)	0.89 (0.88−0.91)	0.86 (0.85−0.88)
SVM	Wilcoxon	0.78 (0.76−0.79)	0.71 (0.69−0.75)	0.8 (0.77−0.82)	0.57 (0.53−0.59)	0.89 (0.88−0.9)	0.86 (0.86−0.87)

Data are given as median and 95% confidence interval. AUC Area under curve at receiver operating characteristics analysis, FSM Feature selection method, LDA Linear discriminant analysis, MI Mutual information, MLP Multilayer perceptron (neuronal network), MRMI Maximum relevance minimum redundancy, NPV Negative predictive value, PLS Partial least squares, PPV Positive predictive value, Rpart Recursive partition, SVM Support vector machine

Back to article page