Incorporation of local dependent reliability information into the Prior Image Constrained Compressed Sensing (PICCS) reconstruction algorithm

详细信息    查看全文

• 作者：Sven Vaegler¹ ; ^{Vaegler_S@ukw.de" class="auth_mail" title="E-mail the corresponding author} ; Dzmitry Stsepankou² ; Jü ; rgen Hesser² ; Otto Sauer¹
• 关键词：CBCT image reconstruction ; prior information ; Compressed Sensing
• 刊名：Zeitschrift für Medizinische Physik
• 出版年：2015
• 出版时间：December 2015
• 年：2015
• 卷：25
• 期：4
• 页码：375-390
• 全文大小：3793 K

文摘

The reduction of dose in cone beam computer tomography (CBCT) arises from the decrease of the tube current for each projection as well as from the reduction of the number of projections. In order to maintain good image quality, sophisticated image reconstruction techniques are required. The Prior Image Constrained Compressed Sensing (PICCS) incorporates prior images into the reconstruction algorithm and outperforms the widespread used Feldkamp-Davis-Kress-algorithm (FDK) when the number of projections is reduced. However, prior images that contain major variations are not appropriately considered so far in PICCS. We therefore propose the partial-PICCS (pPICCS) algorithm. This framework is a problem-specific extension of PICCS and enables the incorporation of the reliability of the prior images additionally.

Material and Methods

We assumed that the prior images are composed of areas with large and small deviations. Accordingly, a weighting matrix considered the assigned areas in the objective function. We applied our algorithm to the problem of image reconstruction from few views by simulations with a computer phantom as well as on clinical CBCT projections from a head-and-neck case. All prior images contained large local variations. The reconstructed images were compared to the reconstruction results by the FDK-algorithm, by Compressed Sensing (CS) and by PICCS. To show the gain of image quality we compared image details with the reference image and used quantitative metrics (root-mean-square error (RMSE), contrast-to-noise-ratio (CNR)).

Results

The pPICCS reconstruction framework yield images with substantially improved quality even when the number of projections was very small. The images contained less streaking, blurring and inaccurately reconstructed structures compared to the images reconstructed by FDK, CS and conventional PICCS. The increased image quality is also reflected in large RMSE differences.

Conclusions

We proposed a modification of the original PICCS algorithm. The pPICCS algorithm incorporates prior images as well as information about location dependent uncertainties of the prior images into the algorithm. The computer phantom and experimental data studies indicate the potential to lowering the radiation dose to the patient due to imaging while maintaining good image quality.