Identifying radiologically significant incidental breast lesions on chest CT: The added value of artificial intelligence
1Department of Radiology, Duke University Medical Center, 2301 Erwin Rd, Durham, NC, 27710, United States.
Related Experiment Videos
View abstract on PubMed
Summary
Artificial intelligence (AI) algorithms improved the detection of radiologically significant incidental breast lesions (RSIBLs) missed by radiologists on chest CT scans. A targeted review workflow efficiently identified these missed lesions.
Area of Science:
- Radiology
- Artificial Intelligence
- Oncology
Background:
- Incidental breast lesions are common findings on chest CT scans.
- Original interpreting radiologists (OIRs) often miss these radiologically significant incidental breast lesions (RSIBLs).
Purpose of the Study:
- To evaluate the efficiency of AI algorithms in detecting RSIBLs missed by OIRs on chest CT examinations.
- To assess the impact of an AI-driven workflow on RSIBL detection rates and workflow efficiency.
Main Methods:
- A retrospective multi-institutional study analyzed chest CT examinations using visual classifier (VC) and natural language processing (NLP) AI algorithms.
- Potential RSIBLs were flagged by AI and reviewed by radiologists; disagreements were adjudicated.
- Statistical analysis compared RSIBLs identified by AI versus OIRs, assessing size and margin differences.
Main Results:
- AI algorithms identified 90.8% of RSIBLs, significantly higher than OIRs (39.5%).
- Missed RSIBLs by OIRs were smaller (1.4 cm) than identified lesions (3.0 cm).
- The AI-based workflow reduced the number of images viewed by 97.3% compared to a full double-read approach.
Conclusions:
- AI-based approaches enhance the detection rates of RSIBLs on chest CT scans.
- Despite an increase in false positives, a targeted review process using AI enables efficient detection of missed RSIBLs.