Two-Material Decomposition From a Single CT Scan Using Statistical Image Reconstruction

Transcription

1 / 5 Two-Material Decomposition From a Single CT Scan Using Statistical Image Reconstruction Yong Long and Jeffrey A. Fessler EECS Department James M. Balter Radiation Oncology Department The University of Michigan AAPM 20 August 3, 20

2 2 / 5 Introduction: DECT Motivation: Provide information about material composition for Radiotherapy, dose calculation and anatomy segmentation PET/CT, attenuation correction Virtual nonenhanced images Popular methods: Dual-energy CT (DECT) Disadvantage: Require two scans or specialized scanners (e.g., fast kvp-switching, dual source-detector CT)

3 Introduction: PWLS Method Using A Single CT Scan 3 / 5 Propose a penalized weighted least-squares (PWLS) method Edge-preserving regularization Reconstruct two basis materials (e.g., soft tissue and bone) Single energy CT scan acquired with X-ray filters Using a split or bow-tie filter Create incident spectra differences among detector channels Require only attachment and alignment of metal filters between the X-ray tube and the patient

4 Split Filtration 4 / 5 R B L x mm Al 200 µm Mo object P I(E) Filter L A Filter R X-ray source fan beam A fan-beam CT scanner with a split filter Energy [kev] Sample spectra at two half filters The effective energies are 49 and 58 kev. [Rutt and Fenster, J. Comp. Assisted Tomo., 980] [Taschereau et al., PMB, 200]

5 Bow-tie Filtration 5 / 5 x 0 0 object γ = 26 γ = 4 γ = 0 γ = 0 I(E) Bow-tie filter X-ray source A fan-beam CT scanner with a bow-tie filter Energy [kev] Sample spectra at four fan angles (γ) The effective energies are 49, 5, 53 and 56 kev.

6 Object Model 6 / 5 µ ( x, E ) = N 2 p ( β l (E) b j x ) x lj l= j= β l (E): the energy-dependent mass attenuation coefficient of the lth material type (e.g., soft tissue and bone) (known) { bj ( x )} : spatial basis functions (e.g., pixels) x lj : density of the lth material at the jth location (unknown)

7 Polyenergetic Measurement Model 7 / 5 ȳ i (x) = I i e f i(s i (x)) + r i noisy measurement ( f i (s i ) = log I i (E) e P ) 2 l= β l(e) s il (x) de s il (x) I i = = N p L i j= I i ( b j x ) x lj dl I i (E) de component line integrals total source intensity i indexes rays and l =, 2 indexes basis materials. Incident intensity I i (E) varies among rays depending on filtration types.

8 Penalized Weighted Least-Squares (PWLS) Reconstruction Logarithm sinogram estimates ˆf i PWLS reconstruction ˆfi = log ( Yi r i I i ) ˆx Ψ(x) = arg minψ(x) x 0 = N d i= ) 2 2 w i (ˆfi f i (s i (x)) + βr(x) where w i = Y i values are statistical weighting factors. 8 / 5

9 NCAT phantom Soft tissue 2 Bone 2 Density The units are physical density (g/cm 3 ) NCAT phantom: [Segars Tsui, IEEE TNS, 2002] 9 / 5

10 Split Filter Results: Soft Tissue Error 0 / 5 JS-FBP PWLS RMS error: g/cm 3 RMS error: g/cm 3

11 Split Filter Results: Bone.6 profiles true 2 density[g/cm 3 ] true JS FBP pwls PWLS produced lower noise, similar edge sharpness. PWLS reduced RMS error from g/cm 3 to g/cm 3. PWLS exhibts 0.03g/cm 3 bias. / 5

12 Split Filter Results: Density JS-FBP.2 PWLS RMS error: g/cm 3 RMS error: g/cm 3 JS-FBP: [Joseph and Spital, J. Comp. Assisted Tomo., 978] PWLS reduced beam-hardening artifacts more effectively 2 / 5

13 Profiles of Attenuation at 70 KeV profiles profiles µ[hu] true JS FBP pwls µ[hu] true JS FBP pwls Split Bow-tie Split and bow-tie filtration methods had similar results. 3 / 5

14 4 / 5 Summary Statistical PWLS method Two basis materials Single energy CT scan Differential filtration creates spectral differences among rays Require only attachment and alignment of metal filters between the X-ray tube and the patient Optimizing materials and thickness or filtration type needed

15 5 / 5 Discussion Inevitable overlap of the filtered spectra Practical issues of using filters Precisely align the filters and rotational center Split filters for tilted rays in 3D CT geometries Adjust radiation dose according to X-ray tube voltages Sensitivity to model mismatch: Compton scatter or imperfect spectral models Investigate choosing regularizers and optimizing their parameters Extend to three material reconstruction using dual-energy CT

16 Split Filter Results: Soft Tissue Images JS-FBP PWLS / 5

17 Split Filter Results: Soft Tissue Profiles true profile density[g/cm 3 ] true JS FBP pwls PWLS produced lower noise, similar edge sharpness. PWLS exhibts 0.05g/cm 3 bias. 7 / 5

18 Split Filter Results: Attenuation at 70 KeV JS-FBP 200 PWLS RMS error: 36 HU RMS error: 8 HU 8 / 5

19 Bow-tie Filter Results: Density JS-FBP.2 PWLS RMS error: g/cm 3 RMS error: g/cm 3 JS-FBP: [Joseph and Spital, J. Comp. Assisted Tomo., 978] PWLS reduced beam-hardening artifacts more effectively 9 / 5

20 Bow-tie Filter Results: Soft Tissue Error 20 / 5 JS-FBP PWLS RMS error: g/cm 3 RMS error: g/cm 3

21 Bow-tie Filter Results: Bone.6 profiles true 2 density[g/cm 3 ] true JS FBP pwls PWLS produced lower noise, similar edge sharpness. PWLS reduced RMS error from g/cm 3 to g/cm 3. PWLS exhibts 0.03g/cm 3 bias. 2 / 5

22 Bow-tie Filter Results: Soft Tissue Images JS-FBP PWLS / 5

23 Bow-tie Filter Results: Soft Tissue Profiles true profile density[g/cm 3 ] true JS FBP pwls PWLS produced lower noise, similar edge sharpness. PWLS exhibts 0.05g/cm 3 bias. 23 / 5

24 Bow-tie Filter Results: Attenuation at 70 KeV JS-FBP 200 PWLS RMS error: 34 HU RMS error: 9 HU 24 / 5