Can Artificial Intelligence Reliably Report Chest X-Rays?: Radiologist\n Validation of an Algorithm trained on 2.3 Million X-Rays

Abstract

Background: Chest X-rays are the most commonly performed, cost-effective\ndiagnostic imaging tests ordered by physicians. A clinically validated AI\nsystem that can reliably separate normals from abnormals can be invaluble\nparticularly in low-resource settings. The aim of this study was to develop and\nvalidate a deep learning system to detect various abnormalities seen on a chest\nX-ray. Methods: A deep learning system was trained on 2.3 million chest X-rays\nand their corresponding radiology reports to identify various abnormalities\nseen on a Chest X-ray. The system was tested against - 1. A three-radiologist\nmajority on an independent, retrospectively collected set of 2000\nX-rays(CQ2000) 2. Radiologist reports on a separate validation set of 100,000\nscans(CQ100k). The primary accuracy measure was area under the ROC curve (AUC),\nestimated separately for each abnormality and for normal versus abnormal scans.\nResults: On the CQ2000 dataset, the deep learning system demonstrated an AUC of\n0.92(CI 0.91-0.94) for detection of abnormal scans, and AUC(CI) of\n0.96(0.94-0.98), 0.96(0.94-0.98), 0.95(0.87-1), 0.95(0.92-0.98),\n0.93(0.90-0.96), 0.89(0.83-0.94), 0.91(0.87-0.96), 0.94(0.93-0.96),\n0.98(0.97-1) for the detection of blunted costophrenic angle, cardiomegaly,\ncavity, consolidation, fibrosis, hilar enlargement, nodule, opacity and pleural\neffusion. The AUCs were similar on the larger CQ100k dataset except for\ndetecting normals where the AUC was 0.86(0.85-0.86). Interpretation: Our study\ndemonstrates that a deep learning algorithm trained on a large, well-labelled\ndataset can accurately detect multiple abnormalities on chest X-rays. As these\nsystems improve in accuracy, applying deep learning to widen the reach of chest\nX-ray interpretation and improve reporting efficiency will add tremendous value\nin radiology workflows and public health screenings globally.\n

Ähnliche Arbeiten