Workshop on Solving Computational Challenges in Medical Imaging

submission for Workshop on Solving Computational Challenges in Medical Imaging – July 2007 Targeted Statistical Image Reconstruction for Cardiac CT Imaging Adam Alessio, PhD, [email protected] Department of Radiology University of Washington Purpose: Current computed tomography (CT) systems generate images with analytic tomographic reconstruction methods because they are simple to implement and fast. These methods can produce high-resolution, low noise images when scans are acquired with a high Xray tube current and patient dose is not a concern. Patient dose has become a major concern for applications such as screening for lung cancer and coronary disease, monitoring response to treatment, and high temporal resolution scans used for coronary CT angiography. Statistical image reconstruction methods, which iteratively solve the tomographic inverse problem, can incorporate the full system response along with information about measurement statistics. These methods allow for reduced patient dose while maintaining image quality and are currently not employed in CT imaging due to computational demands of large CT data and images. We propose a method for reducing the computational demand of statistical reconstruction in cardiac gated studies leading to reduced image noise levels and the potential for reduced patient dose. Methods: The proposed method first performs a single analytic reconstruction of the portion of the image that is stationary using measurements from all cardiac phases. Then, the targeted region of the heart is reconstructed at each phase with a statistical method that only requires forward and backprojections through the target region as opposed to the entire torso. We test the method with simulations of CT measurements derived from patient studies. Results: Simulations of CT angiography scans demonstrate that the targeted reconstruction method is 4x faster and resulted in images with 26% less noise in the cardiac region than conventional statistical reconstruction. Initials studies, with non-optimized image noise models in the statistical algorithm, have shown a 30% reduction in noise compared to standard analytic reconstruction. Conclusion: The proposed targeted method has noise reduction benefits from a) using a statistical reconstruction algorithm and b) constraining a portion of the images with stationary information from all phases. Reconstructing only a targeted ROI offers a clinically viable algorithm for generating low-noise, or potential lower dose, cardiac CT images. Clinical Relevance/Application: This targeted statistical image reconstruction method can improve the trade-off between noise and patient radiation dose with a clinically feasible computation time; dose reduction is particularly important for cardiac CT imaging.