Recent development of Multi-detector Computerized Tomography (MDCT) holds the promise of a non-invasive diagnostic method to evaluate coronary artery disease as an alternative to standard angiography. A major impediment preventing its utilization in routine clinical practice is the presence of image "blooming" artifacts due to vascular calcium. For high-risk patients with serious coronary stenosis who often have significant calcified plaque, this problem becomes very acute. In this paper, we examine an approach to ameliorate the degradations due to the presence of calcium and thus increase the applicability of MDCT. We assume the blooming effect is well modeled as a linear convolutional blur. We then use this model as the basis to perform image restoration, thus deblurring the original image. The aim is to produce CT images with reduced calcium artifacts. We evaluate our method on simulated, ex vivo and clinical data and show that this method has the potential to improve coronary CT image and thus improve stenosis diagnosis.