Nonlinear regression-based method for pseudoenhancement correction in CT colonography - PubMed (original) (raw)
Nonlinear regression-based method for pseudoenhancement correction in CT colonography
Baigalmaa Tsagaan et al. Med Phys. 2009 Aug.
Abstract
In CT colonography (CTC), orally administered positive-contrast tagging agents are often used for differentiating residual bowel contents from native colonic structures. However, tagged materials can sometimes hyperattenuate observed CT numbers of their adjacent untagged materials. Such pseudoenhancement complicates the differentiation of colonic soft-tissue structures from tagged materials, because pseudoenhanced colonic structures may have CT numbers that are similar to those of tagged materials. The authors developed a nonlinear regression-based (NLRB) method for performing a local image-based pseudoenhancement correction of CTC data. To calibrate the correction parameters, the CT data of an anthropomorphic reference phantom were correlated with those of partially tagged phantoms. The CTC data were registered spatially by use of an adaptive multiresolution method, and untagged and tagged partial-volume soft-tissue surfaces were correlated by use of a virtual tagging scheme. The NLRB method was then optimized to minimize the difference in the CT numbers of soft-tissue regions between the untagged and tagged phantom CTC data by use of the Nelder-Mead downhill simplex method. To validate the method, the CT numbers of untagged regions were compared with those of registered pseudoenhanced phantom regions before and after the correction. The CT numbers were significantly different before performing the correction (p<0.01), whereas, after the correction, the difference between the CT numbers was not significant. The effect of the correction was also tested on the size measurement of polyps that were covered by tagging in phantoms and in clinical cases. In phantom cases, before the correction, the diameters of 12 simulated polyps submerged in tagged fluids that were measured in a soft-tissue CT display were significantly different from those measured in an untagged phantom (p<0.01), whereas after the correction the difference was not significant. In clinical cases, before the correction, the diameters of 29 colonoscopy-confirmed 3-14 mm polyps affected by tagging that were measured in a soft-tissue CT display were significantly different from those measured in a lung CT display (p<0.0001) or in colonoscopy (p<0.05), whereas after the correction the difference was not significant. Finally, the effect of the correction was tested on automated detection of 25 polyps > or =6 mm affected by tagging in 56 clinical CTC cases. The application of the correction increased the detection accuracy from 60% with 5.0 FP detections per patient without correction to 96% with 2.9 FP detections with correction. This improvement in detection accuracy was statistically significant (p<0.05). The results indicate that the proposed NLRB method can yield an accurate pseudoenhancement correction with potentially significant benefits in clinical CTC examinations.
Figures
Figure 1
Two examples of the nonlinear characteristic of pseudoenhancement effect. In the images on the left, small circle indicates the location of a sampled CT number in a simulated polyp and a haustral fold, respectively, in an empty phantom. The bar plots in the images on the right indicate the CT number at that location in the empty phantom (white bar) and in tagged phantoms filled with 300, 600, and 900 HU tagging fluids. A nonlinear exponential model (solid curve) fits these measurements more precisely than a linear model (dotted line).
Figure 2
Overview of the study.
Figure 3
Two examples of spatial registration. The crosshairs in the images indicate the same spatial (x,y,z) location within the phantoms. (a) The ROI of untagged reference phantom. (b) The ROIs of two different tagged phantoms before the registration. (c) The ROIs of the two tagged phantoms after the registration.
Figure 4
An example of the ringlike artifact around polyps (the partial-volume region between white and black arrows) that occurs when the optimization algorithm attempts to match uncorrelated partial-volume interfaces of untagged and tagged phantoms.
Figure 5
Application of virtual tagging. (a) A transverse plane of the original untagged reference phantom. (b) A physically tagged phantom. (c) Virtual tagging has been applied to the untagged phantom. (d) The virtually tagged reference phantom has been correlated with the tagged phantom. The white line along the wall of the colon phantom indicates the partial-volume region that has been reconstructed by the application of virtual tagging. The regions that are not included in the optimization are indicated as air (<−900 HU; black).
Figure 6
Two examples of failed correction with the unconstrained objective function of Eq. 7. In (a), soft-tissue region has been corrected excessively. In (b), tagged region has been corrected excessively.
Figure 7
Plots of the regularization terms EST(v¯ST) of Eq. 8 and Etag(v¯tag) of Eq. 9 as a function of CT numbers v¯ST and v¯tag, respectively. The term EST(v¯ST) that is shown in (a) constrains the correction to maintain ST values above −200 HU in the phantoms. The term Etag(v¯tag) that is shown in (b) constrains the correction to maintain tagged values above 200 HU.
Figure 8
Plots of the optimized parameter functions. (a) σ1(⋅). (b) σ2(⋅).
Figure 9
Examples of the average correction of CT numbers within the colonic wall, in a 14 mm polyp, and in a 25 mm polyp. Horizontal axis indicates the parameters of the experiments, whereas the vertical axis represents the average correction in Hounsfield units.
Figure 10
Transverse images of simulated (a) 18 mm and (b) 10 mm polyps and (c) a 3 mm thin fold submerged in 900 HU tagging solution before and after pseudoenhancement correction. Top row: Before correction in LCTD window. Middle row: Before correction in STCTD window. Bottom row: After correction in STCTD window.
Figure 11
Two examples of the application of pseudoenhancement correction to clinical cases with (a) moderately (700–800 HU) and (b) highly radiodense (800–1300 HU) tagged fluid. In both examples, the left figures show a polyp (10 and 6 mm, respectively; arrows) before correction and the right figures after the correction.
Figure 12
Free-response receiver operating characteristic curves of the accuracy of automated detection for clinically significant polyps covered partially or completely by tagged material before (dotted line) and after (solid line) the pseudoenhancement correction. The improvement in detection accuracy by the application of pseudoenhancement correction is statistically significant (p<0.05).
Similar articles
- Scale-based scatter correction for computer-aided polyp detection in CT colonography.
Liu J, Yao J, Summers RM. Liu J, et al. Med Phys. 2008 Dec;35(12):5664-71. doi: 10.1118/1.3013552. Med Phys. 2008. PMID: 19175123 Free PMC article. - Adaptive correction of the pseudo-enhancement of CT attenuation for fecal-tagging CT colonography.
Näppi J, Yoshida H. Näppi J, et al. Med Image Anal. 2008 Aug;12(4):413-426. doi: 10.1016/j.media.2008.01.001. Epub 2008 Jan 26. Med Image Anal. 2008. PMID: 18313349 Free PMC article. - Structure-analysis method for electronic cleansing in cathartic and noncathartic CT colonography.
Cai W, Zalis ME, Näppi J, Harris GJ, Yoshida H. Cai W, et al. Med Phys. 2008 Jul;35(7):3259-77. doi: 10.1118/1.2936413. Med Phys. 2008. PMID: 18697551 Free PMC article. - Fundamental elements for successful performance of CT colonography (virtual colonoscopy).
Park SH, Yee J, Kim SH, Kim YH. Park SH, et al. Korean J Radiol. 2007 Jul-Aug;8(4):264-75. doi: 10.3348/kjr.2007.8.4.264. Korean J Radiol. 2007. PMID: 17673837 Free PMC article. Review. - CT colonography (virtual colonoscopy) for the detection of colorectal polyps and neoplasms. current status and future developments.
Gluecker TM, Fletcher JG. Gluecker TM, et al. Eur J Cancer. 2002 Nov;38(16):2070-8. doi: 10.1016/s0959-8049(02)00384-2. Eur J Cancer. 2002. PMID: 12387832 Review.
Cited by
- Electronic cleansing in computed tomography colonography using AT layer identification with integration of gradient directional second derivative and material fraction model.
Chunhapongpipat K, Boonklurb R, Chaopathomkul B, Sirisup S, Lipikorn R. Chunhapongpipat K, et al. BMC Med Imaging. 2017 Sep 4;17(1):53. doi: 10.1186/s12880-017-0224-6. BMC Med Imaging. 2017. PMID: 28870147 Free PMC article. - Progress in Fully Automated Abdominal CT Interpretation.
Summers RM. Summers RM. AJR Am J Roentgenol. 2016 Jul;207(1):67-79. doi: 10.2214/AJR.15.15996. Epub 2016 Apr 21. AJR Am J Roentgenol. 2016. PMID: 27101207 Free PMC article. Review. - Information-Preserving Pseudo-Enhancement Correction for Non-Cathartic Low-Dose Dual-Energy CT Colonography.
Näppi JJ, Tachibana R, Regge D, Yoshida H. Näppi JJ, et al. Abdom Imaging (2014). 2014 Sep;8676:159-168. doi: 10.1007/978-3-319-13692-9_15. Abdom Imaging (2014). 2014. PMID: 26236780 Free PMC article. - Improving the accuracy of CTC interpretation: computer-aided detection.
Summers RM. Summers RM. Gastrointest Endosc Clin N Am. 2010 Apr;20(2):245-57. doi: 10.1016/j.giec.2010.02.004. Gastrointest Endosc Clin N Am. 2010. PMID: 20451814 Free PMC article. Review.
References
Publication types
MeSH terms
LinkOut - more resources
Full Text Sources