Planning CT‐guided robust and fast cone‐beam CT scatter correction using a local filtration technique
Hehe Cui,Xiao Jiang,Chengyijue Fang,Lei Zhu,Yidong Yang
DOI: https://doi.org/10.1002/mp.15299
IF: 4.506
2021-10-26
Medical Physics
Abstract:PURPOSE: Cone-beam CT (CBCT) has been widely utilized in image-guided radiotherapy. Planning CT (pCT)-aided CBCT scatter correction could further enhance image quality and extend CBCT application to dose calculation and adaptive planning. Nevertheless, existing pCT-based approaches demand accurate registration between pCT and CBCT, leading to limited imaging performance and increased computational cost when large anatomical discrepancies exist. In this work, we proposed a robust and fast CBCT scatter correction method using local filtration technique and rigid registration between pCT and CBCT (LF-RR).METHODS: First of all, the pCT was rigidly registered with CBCT, then forward projection was performed on registered pCT to create scatter-free projections. The raw scatter signals were obtained via subtracting the scatter-free projections from the measured CBCT projections. Based on frequency and intensity threshold criteria, reliable scatter signals were selected from the raw scatter signals, and further filtered for global scatter estimation via local filtration technique. Finally, corrected CBCT was reconstructed with the projections generated by subtracting the scatter estimation from the raw CBCT projections using FDK algorithm. The LF-RR method was evaluated via comparison with another pCT-based scatter correction method based on Median and Gaussian filters (MG method).RESULTS: Proposed method was first validated on an anthropomorphic pelvis phantom, and showed satisfied performance on scatter removal even when anatomical mismatches were intentionally created on pCT. The quantitative analysis was further performed on four clinical CBCT images. Compared with the uncorrected CBCT, CBCT corrected by MG with rigid registration (MG-RR), MG with deformable registration (MG-DR), and LF-RR reduced the CT number error from <math><mrow><mn>79</mn> <mo>±</mo> <mn>35</mn></mrow> </math> to <math><mrow><mn>25</mn> <mo>±</mo> <mn>18</mn></mrow> </math> , <math><mrow><mspace></mspace> <mn>17</mn> <mo>±</mo> <mn>13</mn></mrow> </math> and <math><mrow><mn>7</mn> <mo>±</mo> <mn>3</mn></mrow> </math> HU for adipose and from <math><mrow><mn>115</mn> <mo>±</mo> <mn>61</mn></mrow> </math> to <math><mrow><mn>36</mn> <mo>±</mo> <mn>22</mn></mrow> </math> , <math><mrow><mspace></mspace> <mn>30</mn> <mo>±</mo> <mn>24</mn></mrow> </math> , <math><mrow><mn>7</mn> <mo>±</mo></mrow> </math> 3 HU for muscle, respectively. After correction, the spatial non-uniformity (SNU) of CBCT corrected with MG-RR, MG-DR and LF-RR was <math><mrow><mn>51</mn> <mo>±</mo> <mn>13</mn></mrow> </math> , <math><mrow><mspace></mspace> <mn>60</mn> <mo>±</mo> <mn>21</mn></mrow> </math> , and <math><mrow><mn>21</mn> <mo>±</mo> <mn>9</mn></mrow> </math> HU for adipose, and <math><mrow><mn>50</mn> <mo>±</mo> <mn>22</mn></mrow> </math> , <math><mrow><mspace></mspace> <mn>57</mn> <mo>±</mo> <mn>41</mn></mrow> </math> , and <math><mrow><mn>25</mn> <mo>±</mo> <mn>6</mn></mrow> </math> HU for muscle, respectively. Meanwhile, the contrast-to-noise ratio (CNR) between muscle and adipose was increased by a factor of 2.70, 2.89 and 2.56, respectively. Using the LF-RR method, the scatter correction of 656 projections can be finished within 10 s and the corrected volumetric images (200 slices) can be obtained within 2 min.CONCLUSION: We developed a fast and robust pCT-based CBCT scatter correction method which exploits the local-filtration technique to promote the accuracy of scatter estimation and is resistant to pCT-to-CBCT registration uncertainties. Both phantom and patient studies showed the superiority of the proposed correction in imaging accuracy and computational efficiency, indicating promisingfuture clinical application.
radiology, nuclear medicine & medical imaging