A Novel Reconstuction Framework of Prior Image Constrained Compressed Sensing (PICCS) Enabling the Use of Prior Images with Major Deviations
S Vaegler1*, D Stsepankou2, J Hesser2, O Sauer1, (1) University of Wuerzburg, Wuerzburg, Bavaria, (2) University Medical Center Mannheim, Mannheim, Baden-WuertembergSU-D-116-1 Sunday 2:05PM - 3:00PM Room: 116
Purpose: To develop and investigate an improved reconstruction framework of Prior Image Constrained Compressed Sensing (PICCS) using prior images with major deviations.
Methods: The novel idea of our reconstruction framework is the incorporation of a weighting matrix into the PICCS algorithm. This modification decomposes the prior image into areas that can be taken as a priori information (certain areas) and those regions that are not reliable due to motion or change in structure for the time series (uncertain areas).
We applied our method to the problem of image reconstruction from few projections. Simulations were performed with the Shepp-Logan-Phantom considering 7 to 20 projections and different noise levels.
Compared to the target image, the prior image contained major variations in shape and position of the interior structures. Objective function minimization was based on the ASD-POCS framework using the Ordered-Subset Simultaneous Algebraic Reconstruction Technique.
For all simulations, an onboard imaging system at a linac was mimic (Source-Detector-Distance=1536mm, Source-Axis-Distance=1000mm, 512 bins detector with an overall length of 409.6mm). The size of the image array was 256x256 pixels with a resolution of 1x1mm². The projections were generated from equidistant angles covering 196°.
Image quality was quantified by the root-mean-squared-error (RMSE).
Results: Our method allowed accurate reconstruction of highly undersampled data. In comparison to the PICCS standard, image quality was severely improved. The increased image quality was clearly visible and is reflected in a decreased RMSE of up to 79% for noiseless and 40% for noisy projections.
Conclusion: Our algorithm based on PICCS, using location dependent uncertainties of the prior image, enabled image reconstruction with increased quality even when prior images with major deviations were used. The concept indicates the potential for dose reduction while maintaining good image quality. Further development concerning registration of the prior image and real data reconstruction is currently in preparation.