Encrypted login | home

Program Information

Reduction of CT Ring Artifacts Using Improved Edge Detection and Fuzzy Neural Network

no image available
Z Chao

Z Chao*, H Lee , S Choi , D Kim , H Kim , Yonsei University, Wounju, Gangwon

Presentations

WE-RAM1-GePD-J(B)-3 (Wednesday, August 2, 2017) 9:30 AM - 10:00 AM Room: Joint Imaging-Therapy ePoster Lounge - B


Purpose: Ring artifacts degrade the quality of CT images, so reducing ring artifacts is a important topic. In post-method domain, ring artifacts are converted into line artifacts by polar coordinates transformation. On this basis, many methods have been applied to suppress line artifacts, but the effects are not obvious. Therefore we proposed a new method by improved edge detection and fuzzy neural network in polar coordinates.

Methods: We detected the image edges by improved canny algorithm(The line artifacts were enhanced by Laplacian operator; Prewitt template was used to replace the original template; Suppressing non-maximum gradients were operated only for line artifacts directions). The line artifacts were then located by pixel connectivity method. A bicubic interpolation was performed to compensate artifact region. Besides, the obtained data were trained by Takagi-Sugen fuzzy and Radial Basis Function Network system. Finally, we acquired the results by inverse coordinates transformation. In this experiment, we used 4 different sections of patients’ head data from a CBCT scanner. The peak signal to noise ratio(PSNR) and structural similarity(SSIM) were used to quantitatively compare wavelet-Fourier method and edge detection method.

Results: The Fourier method and edge detection method generated worse results, where the eliminations of artifacts were not realistic by showing lower PSNR and SSIM values. On the contrary, ring artifacts reduction based on our new method presented much better results with the highest performance PSNR and SSIM.

Conclusion: Our new method can reduce ring artifacts better than the other two methods, and fuzzy neural network can improve the performance of interpolation for removing the ring artifacts.


Contact Email: