AI-Enhanced Dual Reading Improves Detection of Missed Chest X-ray Diagnoses

Overview

This study evaluated an AI-assisted double reading system for chest radiographs across two centers, demonstrating improved identification of clinically significant missed findings. The AI system analyzed images and reports to flag discrepancies, facilitating targeted radiologist review without disrupting workflow.

Background

Chest radiography is a widely used, first-line imaging modality for cardiothoracic diseases but is prone to interpretation errors due to its two-dimensional nature and blind spots. Missed findings such as lung nodules, pneumonia, and pneumothorax can lead to delayed diagnoses and adverse patient outcomes. Double reading by peers reduces errors but is resource-intensive, and AI tools have shown promise in improving detection sensitivity. However, integrating AI into routine workflows remains challenging due to potential increases in reading time and false positives. An AI-assisted double reading approach may mitigate these issues by reviewing reports post-interpretation to identify discrepancies efficiently.

Data Highlights

The study retrospectively analyzed chest radiographs from two institutions: a secondary care hospital and a tertiary oncology center. The AI software focused on clinically actionable missed findings including nodules/masses, consolidation, pneumothorax, pneumomediastinum, pneumoperitoneum, rib fractures, and device malposition. Only posteroanterior (PA) chest radiographs were included. The AI system combined deep learning image analysis with natural language processing of radiology reports to detect discrepancies, triggering radiologist review of flagged cases.

Key Findings

• The AI-assisted double reading system successfully identified clinically relevant missed findings on chest radiographs in both general and oncology settings.
• Key pathologies detected included lung nodules/masses, consolidation, pneumothorax, and device malposition, which are critical for patient management.
• The approach did not interrupt the primary radiologist workflow, as AI review occurred after report finalization.
• Only a subset of examinations with AI-identified discrepancies required additional radiologist review, optimizing resource use.
• The AI software provided heatmaps for explainability, aiding radiologists in verifying flagged abnormalities.

Clinical Implications

Implementing AI-assisted double reading can enhance detection of missed clinically significant findings on chest radiographs without increasing radiologist workload during initial interpretation. This targeted review approach may improve diagnostic accuracy and patient outcomes while maintaining efficiency in high-volume imaging settings. Integration of explainable AI outputs supports radiologist trust and verification.

Conclusion

AI-enhanced double reading of chest radiographs is a feasible and effective strategy to reduce missed diagnoses across diverse clinical environments. This method balances improved detection with workflow efficiency, supporting broader adoption of AI in radiology practice.

References

1. Chest radiography usage and challenges [1-2]
2. Miss rate and diagnostic errors in chest radiography [3-7]
3. Double reading and AI applications in radiology [10-21]
4. AI workflow integration and efficiency studies [22-28]