Fig. 2: Comparisons of model performance and different radiologists.
From: Automated abnormality classification of chest radiographs using deep convolutional neural networks
a Performance of different CNN architectures with different input image sizes on the NIH “ChestX-ray 14” dataset. CNN weights were initialized from the ImageNet pre-trained models. Performances are not significantly different among different input image sizes. The error bars represent the standard deviations to the mean values. b True positive rate (sensitivity) and false positive rate (1-specificity) of different radiologists (#1, #2, #3, and #4) against different ground-truth labels. Left depicts performance comparisons when setting the consensus of radiologists as ground-truth. Right depicts comparisons when setting labels from attending radiologist as ground-truth. AR attending radiologist, CR consensus of radiologists (vote by the majority of three board-certified radiologists), AI the artificial intelligence model (ResNet18 CNN model shown here).
