Table 5 Performance of CNN combined with different classification models for the diagnosis of H. pylori infection by single endoscopic image from gastric antrum.

Method	Accuracy	Sensitivity	Specificity	PPV	NPV	AUC
CNN	0.80	0.93	0.66	0.77	0.88	0.79
CNN-KNN	0.84	0.88	0.80^a	0.84	0.84	0.84
CNN-SVM	0.83	0.86	0.78^b	0.83	0.82	0.82
CNN-RF	0.83	0.85	0.81^c	0.85	0.82	0.83
CNN-GBDT	0.78	0.82	0.73	0.78	0.77	0.77
CNN-AdaBoost	0.83	0.87	0.79^d	0.83	0.83	0.83
CNN-XGBoost	0.87	0.88	0.86^e	0.89	0.86	0.87
CNN-LGBoost	0.87	0.89	0.86^f	0.89	0.86	0.87
CNN-CatBoost	0.86	0.89	0.84^g	0.87	0.86	0.86

CNN, KNN, SVM, RF, GBDT, AdaBoost, XGBoost, LGBoost, CatBoost, PPV, NPV and AUC are short for Convolutional Neural Networks, K-Nearest Neighbor, Support Vector Machine, Random Forest, Gradient Boosting Decision Tree, Adaptive Boosting, eXtreme Gradient Boosting, Light Gradient Boosting, Categorical Boosting, Positive Predictive Value, Negative Predictive Value and Area Under the ROC curve. ^aP = 0.002 for CNN-KNN vs CNN; ^bP = 0.008 for CNN-SVM vs CNN; ^cP = 0.001 for CNN-RF vs CNN; ^dP = 0.004 for CNN-AdaBoost vs CNN; ^eP < 0.001 for CNN-XGBoost vs CNN; ^fP < 0.001 for CNN-LGBoost vs CNN; P < 0.001 for CNN-CatBoost vs CNN.

Quick links

Search