The objective evaluation of the phase image: a comparison of different automated methods

One hundred and twenty eight patients with suspected or proven CAD were investigated using both X-ray ventriculography and equilibrium gated radionuclide angiography. In order to diagnose regional wall motion abnormalities, the parametric images obtained by Fourier analysis of the radionuclide images were analysed by different automated methods based on the measurement of the homogeneity of the phase values within the LV ROI. The effect of a diastolic frames exclusion, smoothing the original data, weighting the phase histogram, using Bacharach's error corrected phase distribution functions, using different descriptors of the spread of the phase histograms or distribution functions were tested. Using the results of the X-ray examination as the gold standard, ROC curves were plotted for each method. The ROC curves were modelled by a binormal model using the maximum likelihood method. Statistical tests were applied on the area under the ROC curves. The results show that the diagnostic value of the automated methods depends mainly on the way the histograms or distribution functions are described and to a lesser extent on the type of histograms or distribution functions used. The best result is obtained after smoothing, diastolic frames exclusion, weighting the phase histogram by the amplitude and describing it by its standard deviation. Nevertheless, this result is not significantly different from the result obtained by visual analysis of the phase and amplitude images.