CheXpert: A Large Chest Radiograph Dataset with Uncertainty Labels and Expert Comparison
Jeremy Irvin Pranav Rajpurkar Affiliation: Department of Computer Science, Stanford University Affiliation: Equal contribution Michael Ko Affiliation: Department of Computer Science, Stanford University Affiliation: Equal contribution Yifan Yu Affiliation: Department of Computer Science, Stanford University Affiliation: Department of Computer Science, Stanford University Silviana Ciurea-Ilcus Chris Chute Affiliation: Department of Computer Science, Stanford University Henrik Marklund Affiliation: Department of Computer Science, Stanford University Behzad Haghgoo Affiliation: Department of Computer Science, Stanford University Affiliation: Department of Computer Science, Stanford University Robyn Ball Katie Shpanskaya Affiliation: Department of Medicine, Stanford University Jayne Seekins Affiliation: Department of Radiology, Stanford University David A. Mong Affiliation: Department of Radiology, Stanford University Affiliation: Department of Radiology, Stanford University Safwan S. Halabi Jesse K. Sandberg Affiliation: Department of Radiology, Stanford University Ricky Jones Affiliation: Department of Radiology, Stanford University David B. Larson Affiliation: Department of Radiology, Stanford University Affiliation: Department of Radiology, Stanford University Curtis P. Langlotz Bhavik N. Patel Affiliation: Department of Radiology, Stanford University Matthew P. Lungren Affiliation: Department of Radiology, Stanford University Andrew Y. Ng Affiliation: Department of Computer Science, Stanford University Affiliation: Department of Radiology, Stanford University Affiliation: Equal contribution{jirvin16, Affiliation: Equal contribution{jirvin16,
Abstract
Large, labeled datasets have driven deep learning methods to achieve expert-level performance on a variety of medical imaging tasks. We present CheXpert, a large dataset that contains 224,316 chest radiographs of 65,240 patients. We design a labeler to automatically detect the presence of 14 observations in radiology reports, capturing uncertainties inherent in radiograph interpretation. We investigate different approaches to using the uncertainty labels for training convolutional neural networks that output the probability of these observations given the available frontal and lateral radiographs. On a validation set of 200 chest radiographic studies which were manually annotated by 3 board-certified radiologists, we find that different uncertainty approaches are useful for different pathologies. We then evaluate our best model on a test set composed of 500 chest radiographic studies annotated by a consensus of 5 board-certified radiologists, and compare the performance of our model to that of 3 additional radiologists in the detection of 5 selected pathologies. On Cardiomegaly, Edema, and Pleural Effusion, the model ROC and PR curves lie above all 3 radiologist operating points. W
原文 arXiv:1901.07031;中英对照 + 大白话阅读 https://aha.fim.ai/paper/1901.07031v1