REFLACX, a dataset of reports and eye-tracking data for localization of abnormalities in chest x-rays

Abstract

Deep learning has shown recent success in classifying anomalies in chest x-rays, but datasets are still small compared to natural image datasets. Supervision of abnormality localization has been shown to improve trained models, partially compensating for dataset sizes. However, explicitly labeling these anomalies requires an expert and is very time-consuming. We propose a method for collecting implicit localization data using an eye tracker to capture gaze locations and a microphone to capture a dictation of a report, imitating the setup of a reading room, and potentially scalable for large datasets. The resulting REFLACX (Reports and Eye-Tracking Data for Localization of Abnormalities in Chest X-rays) dataset was labeled by five radiologists and contains 3,032 synchronized sets of eye-tracking data and timestamped report transcriptions. We also provide bounding boxes around lungs and heart and validation labels consisting of ellipses localizing abnormalities and image-level labels. Furthermore, a small subset of the data contains readings from all radiologists, allowing for the calculation of inter-rater scores.

Self-Supervised Pretraining Enables High-Performance Chest X-Ray Interpretation Across Clinical Distributions

“…To assess whether models can generalize across clinical distributions, we chose a wide variety of downstream CXR datasets that we used to finetune and validate our models. These datasets came from diverse sites, including Brazil 23 , China 24,25 , United States ,25,,26,27,28,29,30,31 , Spain 32 , Japan 33 , Vietnam 34 and other countries in Eastern Europe and Central Asia. 35 They also addressed a large set of tasks, such as lung nodule detection, line and tube placement, edema severity classification, pediatric pneumonia detection, pneumothorax detection and multi-class differential diagnosis; some datasets address wide-ranging detection tasks for multiple pathologies.…”