Related Experiment Video
Updated: Jul 3, 2025
07:53
Author Spotlight: Advancing 3D Modeling for Enhanced Diagnosis and Treatment of Pulmonary Nodules in Early-Stage Lung Cancer
Published on: October 13, 2023
1.4K
A deep-learning-based framework for identifying and localizing multiple abnormalities and assessing cardiomegaly in
Weijie Fan1, Yi Yang2, Jing Qi2
1Department of Radiology, Second Affiliated Hospital, Army Medical University, Chongqing, 400037, P. R. China.
Nature Communications
|February 14, 2024
Summary
A new deep learning framework accurately identifies and localizes 14 chest X-ray abnormalities and calculates cardiothoracic ratio. This AI tool surpasses senior radiologists, offering excellent clinical applicability.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Radiology
Background:
- Accurate identification and localization of chest X-ray (CXR) abnormalities are critical for diagnosis.
- Limited availability of large, annotated CXR datasets hinders deep learning-based localization research.
Purpose of the Study:
- To develop a deep learning framework for simultaneous identification and localization of 14 common CXR abnormalities.
- To create a large-scale CXR dataset (CXR-AL14) with bounding box annotations for training and validation.
- To assess the framework's performance in calculating the cardiothoracic ratio (CTR).
Main Methods:
- Construction of the CXR-AL14 dataset, comprising 165,988 CXRs and 253,844 bounding boxes.
- Development of a deep learning framework for abnormality identification, localization, and CTR calculation.
- Rigorous evaluation on held-out, multicentre, and prospective test datasets.
Main Results:
- The framework achieved mean average precision values of 0.572-0.631 for 14 abnormalities at an IoU threshold of 0.5.
- The CTR algorithm demonstrated an intraclass correlation coefficient exceeding 0.95.
- The AI framework exhibited performance superior to senior radiologists in clinical settings.
Conclusions:
- The developed deep learning framework demonstrates excellent performance, generalization, and clinical applicability for CXR interpretation.
- The CXR-AL14 dataset facilitates advanced deep learning research in CXR abnormality localization.
- This AI tool is suitable for routine clinical use, enhancing diagnostic accuracy and efficiency.
