Yinsheng Li 1, Peter Bannas 2, M.D., Perry Pickhardt M.D. 2, Meghan Lubner M.D. 2, Ke Li Ph.D. 1,2, and Guang-Hong Chen Ph.D. 1,2

Transcription

1 Yinsheng Li 1, Peter Bannas 2, M.D., Perry Pickhardt M.D. 2, Meghan Lubner M.D. 2, Ke Li Ph.D. 1,2, and Guang-Hong Chen Ph.D. 1,2 1. Department of Medical Physics, University of Wisconsin-Madison 2. Department of Radiology, University of Wisconsin-Madison

2 Guang-Hong Chen PhD Research funded by GE Healthcare, Siemens AG Royalty recipient from GE Healthcare Other people - Nothing to disclose. 2

3 Metal artifacts often contaminate CT images and hinder medical diagnoses around the metal implant and surrounding soft tissue. The required use of raw data for many artifact reduction methods limits their clinical usage. The purpose of this study was to develop an image domain method to remove metal artifacts directly from clinical CT images. 3

4 * Image segmentation is used for two things: To sort out the metal trace and non-metal contaminated portion in the sinogram To generate a prior image A forward projection of the prior image is used to normalize projection data before a linear interpolation and then de-normalize the projection data after that interpolation FBP reconstruction is performed after the data interpolation *E. Meyer, R. Raupach, M. Lell, B. Schmidt and M. Kachelrieβ. Medical Physics, vol 37, (5482),

5 The introduction of our proposed to the NMAR framework offers two innovations to improve the performance of the NMAR method: DR-PICCS was used to generate the prior image for both normalization and de-normalization; DR-PICCS was used to reconstruct the final image after data interpolation was performed. 5

6 PICCS * Limited view angle range problem Few view problem Noise/dose reduction Artifacts reduction Cardiac CT (TRI-PICCS) Respiratory gated CBCT in IGRT CT perfusion Metal artifacts Reduction () Time-resolved interventional CT Cardiac gated CBCT C-arm CBCT perfusion Dual energy CT *G.-H. Chen, J. Tang, and S. Leng, Med. Phys. (2008) Vol. 35 p660. General X-Ray application (DR-PICCS) 6

7 The proposed method utilizes DR-PICCS* to generate a prior image with bone and background soft tissue preservation; This prior image was then used to correct the metal contaminated image. Comparison with NMAR: The novel prior image was compared with the prior image in NMAR. Corrected images using were compared with those using NMAR. Synthesized projections were used in both methods. Clinical cohort studies: Total of 50 abdominal exam cases: 26 uni-lateral and 24 bilateral Quantitative measurements: mean HU value and standard deviation (STD) Qualitative assessment: two readers (with 15 and 7 years of experience reading abdominal CT) *J. Tang, P.T.,Lauzier and G.H. Chen. Proc SPIE Medical Imaging

8 Image Prior FBP NMAR Display range: [ ] 8

9 Image Prior FBP NMAR Display range: [ ] 9

10 FBP NMAR Display range:[ ] 10

11 FBP Display range:[ ] 11

12 FBP FBP 12

13 FBP Display range:[ ] 13

14 FBP Display range:[ ] 14

15 FBP FBP 15

16 FBP Difference image [ ] Display range:[ ] 16

17 Sagittal Coronal FBP Display range:[ ] 17

18 FBP 18

19 FBP S: most severe streaks M: muscle streaks F: fat streaks *: colon streaks M: muscle reference F: fat reference *: colon reference 19

20 Attenuation (HU) 100 Mean of HU values: FBP vs most severe streak Fat streak Muscle streak Fat reference Muscle reference FBP 20

21 Attenuation (HU) 210 SD of HU values: FBP vs most severe streak Fat streak Muscle streak Fat reference Muscle reference FBP 21

22 Grade Note, a low score represents a high quality image and a high score represents a poor quality image 4 Subjective Image Grading FBP vs D Streak Grading 2D Diagnostic image quality 3D Streak Grading 3D Diagnostic image quality FBP 22

23 Grade 3 (FBP ) Grade 1 () Grade 4 (FBP) Grade 2 () 23

24 The algorithm can be applied to any clinical CT images to reduce metal artifacts This enables clear visualization of both bony structures and surrounding soft tissues. We have found it to be robust for a variety of metal implants including uni-lateral and bilateral metal implants and other cases with multiple metal inserts. Processing time: ~ 1 sec per image slice with an nvidia GTX Titan Z GPU. The total time needed for a volume can be reduced further with multiple GPUs as each slice can be processed independently. 24

25 The work is partially supported by GE Healthcare via a research grant. Thank You e 25

26 Attenuation (HU) Attenuation (HU) Mean of HU values: FBP vs. Air colon streak Air colon reference Air outside body streak Air outside body reference SD of HU values: FBP vs FBP 0 Air colon streak FBP Air colon reference Air outside body streak Air outside body reference 26

27 rsal (%) 120 Relative Streak artifacts level (rsal): FBP vs over different slices over different cases FBP 27

28 Image Prior [ ] NMAR 28

29 Image Prior [ ] NMAR 29

30 FBP Difference image [ ] Display range:[ ] 30

31 FBP Display range:[ ] 31