Abstract PS3-31: Training multi-professional mammogram readers to interpret abbreviated breast MRI (FAST MRI): A UK multi-centre study

Abstract

Abstract Background: As a breast cancer screening test, FAST MRI has the potential to combine the sensitivity of dynamic contrast enhanced breast MRI (DCE-MRI) with costs nearer to those of mammography. In the UK the NHS Breast Screening Programme (NHSBSP) screens women aged 50-70 with a mammogram every 3 years and has a workforce of radiologists, breast clinicians and advanced practitioner radiographers who interpret over 2 million screening mammograms annually. For FAST MRI to be used as a UK population screening tool it will need to be interpreted by the existing multi-professional workforce of mammogram readers. Our research team developed a single study day of standardised training in FAST MRI interpretation. In our single centre pilot study, a group of readers with no previous experience of DCE-MRI interpretation achieved an accuracy just 5% less than expert DCE-MRI readers (Br J Radiol 2019:92 DOI:10.1259/bjr.20190663). Our current study aimed to validate the feasibility of training the NHSBSP workforce in FAST MRI interpretation. We developed display software (MedXViewer) containing ground truth information for each FAST MRI case to train and assess mammogram readers, who represented multiple NHSBSP sites in the South West Region of England. Methods: The study was funded by the National Institute for Health Research (Research for Patient Benefit funding stream; ISRCTN 16624917). A per breast analysis of the frequency of results against the true outcome was obtained overall and for each reader. Differences in accuracy, sensitivity and specificity across reader groups (group 1 = mammogram readers experienced in DCE-MRI interpretation; group 2 = mammogram readers with no previous experience in DCE-MRI interpretation) were analysed using a multilevel generalised mixed model to account for multiple readers per case. Results: 37 NHSBSP mammogram readers at 6 NHSBSP sites attended the training day and completed interpretation of the summative assessment dataset (17 in group 1 and 20 in group 2). All 37 readers completed the reading task of 125 cases (250 breasts), for a total of 9250 reads. Median interpretation time was 99.08 seconds (interquartile range 66.83-150.54). The table below shows per breast analysis comparing the readers’ classification with the true outcome (cancer or normal): The concordance with the ground truth (accuracy) of 83% achieved by Group 2 (4129/5000 (95% CI 82-84%)) was significantly lower than that achieved by Group 1 (3814/4250 (90%; 89-91%); p<0.0001) but differed by only 7%. The concordance with the true outcome improved for the group 2 readers from the first 55 cases to the remaining 70 cases (p=0.02), whereas there was no significant improvement for the expert readers of group 1 (p=0.81). Conclusions: This study validates the feasibility of training the workforce of mammogram readers to interpret FAST MRI with only one day of standardised training. Improvement in performance with experience by the group 2 readers indicates a learning curve and suggests the performance gap between the two groups might be narrowed by further training for group 2. TotalGroup 1Group 2MeasureConcordance (Accuracy)7943/9250 (86%)3814/4250 (90%)4129/5000 (83%)True positive rate (Sensitivity)1806/2109 (86%)858/969 (89%)948/1140 (83%)True negative rate (Specificity)6137/7141 (86%)2956/3281 (90%)3181/3860 (82%)False positive rate1004/7141 (14%)325/3281 (10%)679/3860 (18%)False negative rate303/2109 (14%)111/969 (11%)192/1140 (17%) Citation Format: Lyn IL Jones, Rebecca Geach, Sam A Harding, Andrea Marshall, Sadie McKeown-Keegan, Sian Taylor-Phillips, Premkumar Elangovan, Mark Halling-Brown, Sarah Vinnicombe, Elizabeth O'Flynn, Hesam Ghiasvand, Claire T Hulme, Janet A Dunn. Training multi-professional mammogram readers to interpret abbreviated breast MRI (FAST MRI): A UK multi-centre study [abstract]. In: Proceedings of the 2020 San Antonio Breast Cancer Virtual Symposium; 2020 Dec 8-11; San Antonio, TX. Philadelphia (PA): AACR; Cancer Res 2021;81(4 Suppl):Abstract nr PS3-31.