Deep Learning for Chest Radiology: A Review

Background

Compared to classical computer-aided analysis, deep learning and in particular deep convolutional neural network demonstrates breakthrough performance in many of the sophisticated chest-imaging analysis tasks, and also enables solving new problems that are infeasible to traditional machine learning.

Recent Findings

Deep learning application for radiology has shown that its performance for triaging adult chest radiography has reached a clinically acceptable level, while lung nodule detection from computed tomography has achieved interobserver variability comparable to experienced human observers, and automatically generating text report for chest radiograph is feasible.

Summary

This article will provide a review of leading and emerging deep-learning-based applications in chest radiology.