Background and aim: Established in 2011, BREAST has been recognized as the official training tool for BreastScreen readers which is accessed by hundreds of Australian clinicians to self-evaluate diagnostic performances on clinically relevant test sets in mammography. This study aims to explore the diagnostic accuracy and ability to localise cancerous lesions of Australian surgical trainees via HD (High Definition) and non-HD Full-Field Digital Mammography (FFDM) test sets. In order to contextualise the results, the performance of the surgical trainees was compared to radiology trainees who were at a similar stage of career.
Methods: There were 66 Australian readers who completed the FFDM BREAST test set. Among them, there were 34 surgical trainees (18 reading mammograms in HD mode and 16 reading the same mammograms in non-HD mode) and 32 radiology trainees (HD mode). The readers were asked to read 20 mammograms (14 normal and 6 cancer) and identify lesions suspected to be cancers on each mammographic projection using the RANZCR Tabar grading. Readers’ performances between two groups of trainees were compared with the truth data and then evaluated in terms of specificity, case sensitivity, lesion sensitivity, ROC and JAFROC via Mann-Whitney U test.
Results: The overall performances of radiology trainees and surgical trainees was statistically significant for specificity, sensitivity and JAFROC (P<0.05), with the radiology trainees having higher specificity (72%-vs-34%;P<0.001) and JAFROC (65%-vs-51%;P=0.009) scores and the surgical trainees having higher sensitivity scores (90%-vs-74%;P=0.025). Comparing the performance of the radiology trainees to surgical trainees when reading the mammograms in HD mode, the radiology trainees had higher specificity (72%-vs-35%;P<0.001), ROC (77%-vs-64%;P=0.08) and JAFROC (65%-vs-51%;P=0.045) scores. No statistically significant differences were observed between the surgical trainees on HD cases and the surgical trainees on non-HD cases.
Conclusions: Radiology trainees performed better than surgical trainees in identifying the normal mammograms and had higher confidence in differentiating the normal and abnormal cases. However, the surgical trainees were better in detecting cancer cases which could be because the surgical trainees had more experience in working with cancer patients than radiologists who often work in the screening environment where the rate of cancer patients among screening women is low. The image resolution did not cause differences in the performances of surgical trainees. However, further work will be needed to improve several diagnostic metrics of both radiology and surgical trainees.