Abstract Archives of the RSNA, 2014
Oliver William Edmund Morrish MSc, Presenter: Nothing to Disclose
Lorraine Tucker, Abstract Co-Author: Nothing to Disclose
Richard Black MS, Abstract Co-Author: Nothing to Disclose
Stephen W. Duffy, Abstract Co-Author: Nothing to Disclose
Fiona Jane Gilbert MD, Abstract Co-Author: Speaker, Bracco Group
Research Grant, GlaxoSmithKline plc
Research Grant, General Electric Company
To compare different methods of assessing breast density in a UK screening population.
Conventional two-view digital mammograms were collected from six centres from 8867 women attending either a screening assessment clinic, or screening for family history. Breast density was assessed by 26 experienced image readers on a visual analogue scale and by two automated volumetric density tools (QuantraTM, Hologic and VolparaTM, Mātakina). Total breast volume, fibroglandular volume and percent volume density for each view were combined logically to produce results for each woman.
There was a good linear correlation (R2=0.95) between Quantra and Volpara for total breast volume with a mean difference of 43.5cm3 (5.04% of the mean result ±0.32%, 2sem). There was less correlation for fibroglandular volume (R2=0.74) and the mean difference was 30.3cm3 (21.19%±0.72%). On average Quantra gave the larger value and the difference tended to increase with increased volume. For the percent volume density the mean difference was 1.61% (16.32%±0.69%) and R2 was 0.65. Comparison of the observer’s scores with the area-based Quantra density yielded a poor correlation (R2=0.31). Exponential correlations of observer’s scores with the volumetric density results gave R2 values of 0.33 and 0.38 for Quantra and Volpara respectively.
There is poor correlation of breast density scored by observers and automated techniques. In part this may be explained by differences in the quality being measured (area vs volume) and image presentation (for presentation vs for processing). However inter-observer variability in the UK may be high as quantitative estimations of breast density are not made routinely. If automated methods are to be used differences in their output, caused by differences in the underlying algorithm, need to be understood.
There are differences in the results of each of the breast density measurement methods. These differences need to be understood if management decisions are based on breast density.
Morrish, O,
Tucker, L,
Black, R,
Duffy, S,
Gilbert, F,
Comparing Breast Density Measurement Techniques in Screening Digital Mammography. Radiological Society of North America 2014 Scientific Assembly and Annual Meeting, - ,Chicago IL.
http://archive.rsna.org/2014/14006101.html