Ethical approval was obtained from Jordan University of Science and Technology and a written consent was obtained from all participated radiologists.
Image test set
A set of 60 digital mammograms (cranio-caudal (CC) and medio-lateral oblique MLO projections) of both breasts were included. The test set contain 20 cases with 21 biopsy proven cancer lesions and 40 cancer free cases confirmed after 2 years follow up.
Images were allocated to specific mammographic densities according to the Australian and New Zealand College of Radiology (RANZCR) synoptic guidelines (19):
1) Low mammographic density cases include RANZCR first level <25% glandular and second level RANZCR (2) 25-50% glandular.
2) High mammographic density cases include RANZCR third level 51-75% glandular and fourth level RANZCR (4) >75% glandular.
Image display conditions
The reading sessions were performed in a room with an ambient lighting of 20 lux10 at the position of the observer, measured with a calibrated photometer. For minimum specular reflection, walls were painted with a matt light grey colour. All mammograms were interpreted on 8 Megapixel RADI FORCE 850 monitors calibrated according to the Digital Imaging and Communications in Medicine (DICOM) part 14 standards.
The participated radiologists freely examined each case and had full access to standard processing techniques such as zooming, windowing and panning without any time limitation. Cases were scored using a 1-5 confidence scale (1: is confident that the image is normal and no location is reported, 5: is completely confidently that the image contains a malignant lesion). Any perceived malignancy was localised.
Data analysis
The following metrics were calculated:
· Sensitivity defined as the proportion of positive images which were correctly identified;
· Specificity defined as the proportion of negative images which were correctly identified;
· Area under the receiver operator characteristics (AUC) curves;
· Location sensitivity defined as the proportion of malignancies that were correctly located.
· Jackknife free-response receiver operator characteristics (JAFROC) figure of metric.
Radiologists were grouped into three categories; all radiologists, over 1000 mammograms annual read and less than 1000 mammograms annual read. Comparison of observer performance between high and low mammographic density cases was analysed using a paired t-test. A significance level of p ≤ 0.05 was set for all comparisons.