To estimate the potential reduction in radiologist workload and time saved by substituting one of the two readers with an AI model in BreastScreen Norway, highlighting the significance of this reduction.
Key Findings:
Mean reading time for mammograms was estimated at 41 seconds per examination, indicating a significant efficiency in the screening process.
Consensus/arbitration time averaged 2:12 minutes per examination, highlighting the resource demands of the current system.
AI could potentially reduce the workload by replacing one radiologist, maintaining diagnostic performance, which is crucial for patient outcomes.
Interpretation:
The integration of AI in mammographic screening could significantly alleviate the workload on radiologists, addressing the shortage of qualified personnel while preserving diagnostic accuracy, though challenges in implementation must be considered.
Limitations:
The study relied on aggregated data, which may not capture individual variability in reading times, potentially affecting the generalizability of the findings.
The effectiveness of AI triage depends on the selection of appropriate risk thresholds, which may vary across contexts and could introduce biases.
Conclusion:
Implementing AI in mammographic screening could lead to substantial time savings for radiologists, enhancing the efficiency of breast cancer detection programs while ensuring that diagnostic accuracy is maintained.
