Fig. 1: Receiver operating characteristic curves (ROCs) for different ImageNet pre-trained CNN architectures versus radiologists on the NIH “ChestX-ray 14” dataset. | npj Digital Medicine

Fig. 1: Receiver operating characteristic curves (ROCs) for different ImageNet pre-trained CNN architectures versus radiologists on the NIH “ChestX-ray 14” dataset.

From: Automated abnormality classification of chest radiographs using deep convolutional neural networks

Fig. 1

a Labels voted by the majority of radiologists as the ground-truth reference standard. b Labels derived from text reports as the ground-truth reference standard. Radiologists’ performance levels are represented as single points (or a cross for attending radiologist who wrote the radiology report). AUC area under the curve.

Back to article page