Image quality evaluation of motion-contaminated calcified plaques in cardiac CT

An automated method for evaluating the image quality of motion-contaminated calcified plaques in non contrast-enhanced cardiac CT was developed. This method consisted of using the rapid phase-correlated ROI (RP-ROI) reconstruction algorithm for generating a 4D series of calcified plaque images, and then extracting phase-correlated dynamic, morphological, and intensity-based features of the plaques. Velocity-based linear regression (VLR), multiple linear regression (MLR), and artificial neural network (ANN) regression models were used to predict plaque motion indices based on these extracted features. Since motion indices were reflective of the amount of motion that plaques underwent at multiple phases throughout the cardiac cycle, these indices represented metrics of image quality with respect to motion artifacts. Motion indices were predicted for sixty calcified plaques placed along the left and right coronary arterial trees in the dynamic NCAT phantom. In a five-fold cross-validation scheme involving independent testing groups of twelve plaques, average repeated measures concordance correlation coefficients for the VLR, MLR, and ANN models were 0.848 +/- 0.022, 0.885 +/- 0.014, and 0.950 +/- 0.009. The motion indices predicted by the ANN regression model also were used to obtain optimal phases for image interpretation of individual plaques.