Optimisation of direct digital mammography with the Jackknife FROC method

3 PURPOSE: The aim of this study was to investigate how the diagnostic accuracy in digital mammography is affected by dose to the breast. MATERIAL AND METHODS: Structures resembling invasive tumours and microcalcifications on x-ray images were positioned in an anthropomorphic breast phantom equivalent to a 50 mm compressed breast containing 50% glandular tissue. The average glandular dose based on phantom exposures using the automatic exposure control of the x-ray system (Siemens Mammomat Novation direct digital mammography) was 1.4 mGy, henceforth referred to as the 100% dose level. Thirty digital images were acquired of the phantom at each of three dose levels: 100%, 50% and 30% of the AEC-level. In each image there were at most 3 lesions present at random locations. Eight observers (3 radiologists and 5 non-radiologists) interpreted these 90 images on a DICOM calibrated 5 mega-pixel mammography monitor. For each image the observers were asked to identify the locations of perceived lesions and to assign an integer rating on a 4 point confidence scale. A viewer was designed to make the evaluation easier and more accurate to perform. Observer performance was evaluated by the jackknife FROC (JAFROC) method. RESULTS: The mean FROC figures-of-merit (θ) for detecting the simulated lesions for the 100%, 50% and 30% dose levels, were 0.69, 0.68 and 0.63, respectively. There was no observed statistical difference in detection accuracy among observers between the 100% and the 50% dose level. The number of detected microcalcifications was almost identical (62.5%) at both dose levels, while the detected simulated invasive tumours was about 1% lower at the 50% level. At the 30% level the number of detected lesions was more than 10% lower both for the simulated invasive tumours and the microcalcifications. CONCLUSION: Within the precision of our measurements, it is possible to reduce the absorbed dose to the breast to half of the currently used dose level without compromising diagnostic accuracy.