To evaluate the impact of workflow-integrated generative AI on radiograph documentation efficiency and clinical accuracy, specifically focusing on peer-reviewed accuracy and textual quality.
Key Findings:
Mean documentation time decreased from 189 seconds to 160 seconds, a 15.5% increase in efficiency.
No statistically significant difference in clinical accuracy or textual quality between model-assisted and non-assisted reports.
Model identified unexpected pneumothorax cases with 73% sensitivity and 99.9% specificity, indicating high reliability in detecting critical conditions.
Interpretation:
The study suggests that generative AI can enhance documentation efficiency without compromising report quality, although further research is needed to assess long-term impacts on productivity and clinician workload.
Limitations:
Observational, nonrandomized design may introduce confounding factors, potentially affecting the validity of the efficiency gains.
Study conducted in a single academic health system, limiting generalizability to other practice settings, including community hospitals and nonacademic centers.
Conclusion:
Initial evidence supports the benefits of AI-assisted draft reporting in radiology, indicating potential for improved clinical workflows.
