Development of imaging methods that utilize polychromatic nature of radiation beams

Abstract

We exploit polychromatic nature of radiation beams for development of radiation imaging schemes. First, we suggested a single scan low dose dual energy CT scheme that uses intentionally hardened X-ray beams to increase the mean energy of X-ray beams. Reciprocating multi-slit filter was used to obtain a partially beam hardened X-ray beams. Reduced image quality of high energy CT due to increased quantum noise was improved via our suggested image reconstruction algorithm, i.e., a gradient magnitude information (GMI) minimization, that exploits gradient information of low energy CT image reconstructed beforehand to save the edges of high energy CT image. A feasibility of the suggested dual energy CT scheme and image reconstruction algorithm was analyzed experimentally with cylinder, head, and pork rib phantoms. Image quality of high energy CT was improved through GMI minimization algorithm. Contrast-to-noise ratio (CNR) of each sample material was increased compared to that of the CT image reconstructed through a conventional image reconstruction algorithm, i.e., adaptive-steepest-descent projection-onto-convex-sets (ASD-POCS). In case of cylinder phantom, for example, the value was increased from 29.8 to 102 for Teflon, from 1.59 to 14.5 for LDPE, from 11.5 to 54.5 for Derlin, from 5.72 to 44.9 for PMP, and from 0.74 to 10.7 for polystyrene. We calibrated coefficients of dual energy material decomposition algorithm by use of calibration phantom composed of acryl and aluminum and applied to material decomposition of head phantom and pork rib. In case of head phantom, we have shown that small anatomic structure inside the phantom was conserved through GM minimization algorithm. Bone and soft tissue can be discriminated through the calibrated dual energy algorithm into aluminum- and acryl- based images. In case of pork rib, we have shown that interface between soft tissues can be saved via the GMI minimization algorithm, and that they could be discriminated one another in a acryl-based material decomposed image as well. Polychromatic nature of proton beams modulated to have a monotonically decreasing depth-dose distribution were used for a fast proton CT scan based on an assumption that they behave similar to photon beams. Proton projection images were acquired using the modulated proton beams and the EBT2 film as an intensity detector. Four electron-density cylinders representing normal soft tissues and bone were used as imaged object and scanned at 40 views that are equally separated over 360°. Digitized film images were converted to water equivalent thickness (WET) by use of an empirically derived conversion curve. For improving the image quality, a deconvolution-based image deblurring with an empirically acquired point-spread function (PSF) was employed. The four iterative algorithms, i.e., ASD-POCS, SM-POCS, SM-EM, and EM-TV, have been developed and applied for proton CT image reconstruction. Electron density phantoms that represent human tissues such as bone, muscle, adipose, and lung, were used for analysis of the algorithms. Objects of higher electron density have been reconstructed more accurately than those of lower density objects. The bone, for example, has been reconstructed within 1% error. EM-based algorithms produced an increased image noise and RMSE as the iteration reaches about 20, while the POCS-based algorithms showed a monotonic convergence with iterations. The ASD-POCS algorithm outperformed the others in terms of CNR, RMSE, and the accuracy of the reconstructed relative stopping power in the region of lung and soft tissues. Although it still seems that the images need to be improved for practical applications to the treatment planning, proton CT imaging by use of the modulated beams in sparse-view sampling has demonstrated its feasibility.