| Summary: | A practical evaluation has been made of a methodology for quantitative evaluation of change in breast parenchymal density by computer analysis of digitised mammograms. It was hypothesised that interval changes in radiographic breast density could be quantified by a method of normalisation to a reference density on the mammograms, namely, the pectoralis major muscle. The bilateral lateral-oblique mammograms of ten patients at a two year interval were selected. These forty mammograms and calibrated test tools films were digitised using a charge-coupled device camera (CCD) and a drum-scanning microdensitometer (DSMD). The DSMD was found to be more suitable for this work due to the highly linear response characteristics, wide latitude of response, uniform and low noise levels, accurate film positioning and the possibility of more rapid image analysis. The CCD had a logarithmic linear response with less wide latitude, less uniform and higher noise levels, and variation introduced due inaccuracy in film positioning. For the DSMD, the difference between the breast region of interest (ROI) pixel value and pectoralis muscle ROI pixel value, was confirmed as the most appropriate for the method of normalisation; for the CCD, the ratio of logarithms was suitable. The importance of very meticulous mammographic technique was highlighted. Difficulties with normalisation were mostly due to non-uniform compression and inadequate inclusion of pectoralis muscle. For the DSMD images, a more rapid 'autotrace' method for outlining the pectoralis muscle ROI was comparable in accuracy to the slower method which took measurements from six small ROIs.
|
|---|
