Machine Learning Augmented Interpretation of Chest X-rays: A Systematic Review

Area of Science:

• Radiology
• Artificial Intelligence
• Medical Imaging

Background:

• Chest X-ray (CXR) interpretation faces limitations, driving the development of machine learning (ML) systems.
• Understanding ML capabilities and limitations is crucial for clinical practice integration.

Purpose of the Study:

• To systematically review ML applications for CXR interpretation.
• To assess the performance and impact of ML tools in facilitating CXR analysis.

Main Methods:

• Systematic literature search for ML algorithms detecting >2 radiographic findings on CXRs (Jan 2020 - Sep 2022).
• Summarized model details, study characteristics, risk of bias, and quality.
• Included 46 studies from an initial retrieval of 2248 articles.

Main Results:

• Published ML models demonstrated strong standalone performance, often equaling or surpassing clinician accuracy.
• ML tools improved clinician performance when used as diagnostic aids.
• Models were trained and validated on large datasets (average 128,662 images).

Conclusions:

• ML systems for CXR interpretation exhibit robust performance and enhance clinician diagnostic capabilities.
• These AI tools show potential for improving radiology workflow efficiency.
• Safe implementation requires clinician expertise and addressing identified limitations.