Vectorial total variation denoising for myocardial blood flow estimation in dynamic CT

Dynamic CT could provide a relatively inexpensive and widely available technique to measure myocardial blood flow (MBF) for evaluating coronary artery disease (CAD). The main limitation to clinical acceptance is the substantial radiation dose. The dose can be reduced by decreasing the photon flux or the number of temporal frames, both resulting in noisy 4D data from which accurate MBF information needs to be extracted. In this work, we investigate two novel image domain denoising techniques based on total variation (TV). These methods are possible generalizations of the Rudin Osher Fatemi denoising model to multiple frames of dynamic CT data. Our first method, scalar total variation (STV), considers the sum of total variation of multiple temporal images as the penalty. Our second method, vectorial total variation (VTV) leverages the fact that images reconstructed from different dynamic frames have a common edge structure. We performed simulation studies of low-dose, dynamic cardiac CT acquisitions (25 mAs, 30 frames) of several flow states (flow = 0.5, 1.0, 2.0, 3.0 ml/g/min) and compared the accuracy of MBF estimates obtained from the proposed STV and VTV methods to a sinogram-domain 4D smoothing technique. Through simulation studies, we found that the proposed VTV algorithm gave us the best MBF estimate for 25 mAs tube current. The VTV method is fast and versatile, offering a pragmatic filtering strategy to improve the quality of dynamic 4D CT images. The STV method did not perform as well as the other two denoising methods.