Image compression in digital mammography: effects on computerized detection of subtle microcalcifications

Our previous receiver operating characteristic (ROC) study indicated that the detection accuracy of microcalcifications by radiologists is significantly reduced if mammograms are digitized at 0.1 mm x 0.1 mm. Our recent study also showed that detection accuracy by computer decreases as the pixel size increases from 0.035 mm x 0.035 mm. It is evident that very large matrix sizes have to be used for digitizing mammograms in order to preserve the information in the image. Efficient compression techniques will be needed to facilitate communication and archiving of digital mammograms. In this study, we evaluated two compression techniques: full frame discrete cosine transform (DCT) with entropy coding and Laplacian pyramid hierarchical coding (LPHC). The dependence of their efficiency on the compression parameters was investigated. The techniques were compared in terms of the trade-off between the bit rate and the detection accuracy of subtle microcalcifications by an automated detection algorithm. The mean-square errors in the reconstructed images were determined and the visual quality of the error images was examined. It was found that with the LPHC method, the highest compression ratio achieved without a significant degradation in the detectability was 3.6:1. The full frame DCT method with entropy coding provided a higher compression efficiency of 9.6:1 at comparable detection accuracy. The mean-square errors did not correlate with the detection accuracy of the microcalcifications. This study demonstrated the importance of determining the quality of the decompressed images by the specific requirements of the task for which the decompressed images are to be used. Further investigation is needed for selection of optimal compression technique for digital mammograms.