[Paper Reading]ChestX-ray8: Hospital-scale Chest X-ray Database and Benchmarks on Weakly-Supervised Classification and Localization of Common Thorax Diseases

Problem Definition

How loosely labeled hospital-size chest X-ray database can be used to facilitate the data-hungry deep learning paradigms in building truly large-scale high precision computer-aided diagnosis (CAD) systems.

Contributions

• Propose a new chest X-ray database “ChestX-ray8” with 108,948 frontal-view X-ray images of 32,717 unique patients with the text-mined eight common disease labels, mined from the text radiological reports via NLP techniques
• Demonstrate that some commonly occurred thoracic diseases can be detected and even spatially-located via a unified weakly-supervised multi-label image classification and disease localization formulation

Method

Construction of Hospital-scale Chest X-ray Database

• Short-list eight common thoracic pathology keywords that are frequently observed and diagnosed, based on radiologists’ feedback
• Search the PACS system to pull out all the related radiological reports (together with images) as target corpus
• Use NLP techniques for detecting the pathology keywords and removal of negation and uncertainty

Common Thoracic Disease Detection and Localization

• Tailor Deep Convolutional Neural Network (DCNN) architectures for weakly-supervised object localization, by considering large image capacity, various multi-label CNN losses and different pooling strategies

Experiments

• Dataset: proposed dataset
• Classification and localization
• Evaluation: (classification) AUCs of ROC curves for multi-label classification / (localization) pathology localization accuracy and the average false positive number

Results

Summary

The authors proposed a large-scale chest X-ray dataset and the way to collect it, together with a CNN model to deal with the classification and localization of 8 common diseases

Reference

Wang, Xiaosong, et al. “Chestx-ray8: Hospital-scale chest x-ray database and benchmarks on weakly-supervised classification and localization of common thorax diseases.” Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. 2017.