Effects of lossy compression on lesion detection: predictions of the nonprewhitening matched filter.

The nonprewhitening matched filter (NPWMF) was used to study the effect of wavelet-based lossy image compression on lesion detectability. Two classes of images, representing lesion-present and lesion-absent cases, were generated by simulation techniques. The performance index, da, of the NPWMF was calculated from the reconstructed images of the two classes and was used as an objective measure for quantifying signal detectability. The effect of compression on detectability was analyzed by defining signal amplitude, signal size, and noise amplitude at a constant da. Since human observer studies directly correlate with the performance index da, the effect of signal parameters on compression ratios could be evaluated. For example, a signal with 7 pixels in diameter and an amplitude of 10 that was compressed at a ratio of 4.6:1, and a signal of the same size with an amplitude of 15 but compressed at a ratio of 24.3:1, both had a da of 4.1, implying identical detectability. Similar results for other combinations of signal and noise parameters, but at constant da, can be used to analyze the effect of compression on detectability without requiring human observer studies.