Table 1 CNN with automatically extracted features performed to the level of a trained radiologist when predicting transplant outcomes

From: Improving prognostic accuracy in lung transplantation using unique features of isolated human lung radiographs

 

XGBoost (Manual labels)

CNN (Automatically extracted features)

P value

Accuracy

56.0 ± 3.7

62.7 ± 1.8

0.0007

AUROC

76.3 ± 2.2

76.1 ± 1.2

0.37

  1. AUROC area under the receiver operating characteristic, XGBoost extreme gradient boosting, CNN convolutional neural network.
Back to article page