Evaluation of GPU-Based CT Reconstruction for Morbidly Obese Patients

The obese population is increasing in the United States. There have been modest improvements in scanner hardware and image processing to address some specific challenges associated with imaging of the morbidly obese patients. However, most legacy CT systems lack capabilities to provide sufficient delivery of image-based diagnosis in this increasing subset of population. One of the most common problems is the projection data truncation in CT imaging due to the massive girths of obese patients. In the past decade, it was proved that the image can be accurately and stably reconstructed from locally truncated projections if certain prior knowledge is known, and this technique is named interior tomogrpahy. To overcome the time-consuming issue of the iterative algorithms, we apply GPU techniques to speed up the reconstruction process. In this paper, we evaluate the GPU-based CT reconstruction algorithms (one analytic algorithm and one iterative reconstruction algorithm) for obese patients with both simulated and real clinical datasets. While the approximate analytic reconstruction algorithm outperforms the iterative reconstruction (IR) algorithm in terms of computational cost, the IR algorithm outperforms the analytic one in terms of image quality especially when the projection data is suffered from patient motion, which can happen when the obese patients are not able to hold a breath during the scan.