Room: Exhibit Hall | Forum 8
Purpose: Reduction of dose has been a critical challenge in perfusion CT since tens of successive images need to be acquired in the procedure to record the time intensity curves (TICs). In this work, we investigate the feasibility and performance of a content-oriented sparse representation (COSR) method for denoising in perfusion CT.
Methods: In the COSR denoising method, a CT image is segmented by thresholding into several content-areas. After being ex-painted smoothly outside its boundary, each content-area is sparsely coded by an atom from the dictionary learnt from the image patches extracted from the corresponding content-area. The regenerated content-areas are finally aggregated to form the denoised CT image. The ability of preserving texture and edges in CT image by the COSR method is investigated with a water phantom simulation. A perfusion CT of ischemic stroke is simulated using a digital brain perfusion phantom in absence and presence of noise. The perfusion CT images denoised by COSR are analyzed and evaluated against the ground truth, noiseless and noisy simulations, as well as their corresponding perfusion maps of CBV, CBF, MTT and TTP.
Results: The simulation of water phantom shows that the proposed COSR method can well preserve the texture and edges in perfusion CT images with the denoising ability comparable to that of the conventional SR method. For the perfusion CT phantom, the COSR method can provide denoised images that are almost identical to the noiseless images while preserving texture and edges, and the resultant perfusion maps are less noisy and approach the perfusion maps calculated from the noiseless images.
Conclusion: As shown in the preliminary results, the COSR method can effectively reduce noise in brain perfusion CT images while preserving texture and edges, and thus may find its clinical utility in perfusion CT for ischemic stroke.