Autonomous chest x-ray image classification, capabilities and prospects: rapid evidence assessment

Area of Science:

• Radiology and Medical Imaging
• Artificial Intelligence in Healthcare
• Clinical Workflow Optimization

Background:

• Chest x-ray radiography (CXR) is a primary screening tool, but manual review leads to significant radiologist workload and potential errors.
• Automating the screening of normal CXR studies is crucial for efficient clinical workflows.

Purpose of the Study:

• To evaluate the effectiveness of artificial intelligence (AI) for autonomous CXR triage.
• To assess the potential of AI in routine clinical practice for radiological screening.

Main Methods:

• A rapid evidence assessment methodology was used, searching major databases from 2019-2025.
• Inclusion criteria focused on large-scale studies of multiple pathologies with English abstracts.
• Meta-analysis of diagnostic performance metrics and quality assessment using QUADAS-2, QUADAS-CAD, and GRADE frameworks.

Main Results:

• 11 studies met inclusion criteria; three used real-world datasets, and three analyzed cohorts >500,000 CXRs.
• AI autonomously triaged 15.0%-99.8% of CXRs (weighted average 42.3%), reaching 54.8% in continuous real-world data flow.
• High sensitivity (97.8%) and specificity (94.8%) were observed, with 55% of studies having low risk of bias.

Conclusions:

• AI systems for autonomous CXR triage are nearing clinical readiness.
• AI can decrease radiologist workload, reduce errors, and lower screening costs.
• Regulatory barriers and a scarcity of real-world clinical trials impede implementation.