Prospective Evaluation of Theoretical Time Savings in Brain MRI Using AI-Guided Identification of Examinations Potentially Not Requiring Contrast
2Department of Radiology, Tel Aviv Sourasky Medical Center (Ichilov), Israel
3Research Institute, Assuta Health Services, Israel
4School of Health Systems Management, Netanya Academic College, Israel
Background
To prospectively evaluate the theoretical time-saving potential of a commercial AI tool for identifying routine brain MRI examinations that may not require contrast, using the final radiology report as the reference standard.
Materials and Methods
This prospective observational local validation study included 350 routine gadolinium-enhanced brain MRI examinations. AI output was used to estimate theoretical scan-time savings if contrast-enhanced imaging were omitted in AI-negative examinations. Final radiology reports served as the reference standard. Outcomes included examination duration, estimated time savings, agreement with the reference standard, and binary diagnostic performance. Headache was the most common referral indication.
Results
Valid examination duration was available for 333 cases, AI-derived time-saving data for 312, and agreement data for 334. Mean examination duration was 25.4 minutes, with an estimated 6.8 minutes saved per examination, corresponding to a 25.5% reduction and 35.5 total hours saved. AI agreed with the radiology report in 278/334 examinations (83.2%). Sensitivity was 78.0%, specificity 84.4%, positive predictive value 51.7%, and negative predictive value 94.7% (p < 0.001). Agreement was higher in stroke (78.6%) and hemorrhage (68.8%) cases and lower in tumor cases (35.7%).
Conclusion
In this prospective local validation study, AI-guided identification of brain MRI examinations that may not require contrast showed potential workflow gains, with an estimated 6.8 minutes saved per examination and a 25% reduction in scan time. High specificity and negative predictive value support further evaluation as a rule-out tool in selected routine brain MRI examinations. Lower agreement in tumor cases highlights the need for indication-specific safeguards and further validation before clinical adoption.