CT Metal Artifact Reduction Algorithm Research Based On Frequency Split

Computed tomography (CT) has become one of the most important auxiliary equipment in medical imaging diagnostic. It measures the cumulative attenuation coefficients of X-ray in all directions through the sectional, and then calculates the distribution of the accumulated attenuation coefficient of each of the sectional, the image is displayed in the last and help clinical doctors do the correct diagnosis for the lesion. In recent years, the major manufacturers at home and abroad have introduced new CT constantly, they have been committed to improve image quality, reduce radiation dose, enhance the speed of scanning and optimizing the process of post-processing, while also improve and optimize the removal of the CT metal artifacts method constantly. Because when the object being scanned, there are a large number of high density metal implant, the attenuation coefficient of X-ray through these metal implants are much higher than other body tissues, thus causing the X-ray sharp attenuation after the role of these metals, resulting the projection data distort and artifacts appears in the final reconstructed image,these artifacts severely affected clinicians correct diagnosis of the lesion, these artifacts known as metal artifacts.Because of the serious impact of metal artifacts on clinical diagnosis and the diversity of its manifestations, which has become a key issue in CT imaging processing research. There were a lot of researches about how to overcome these artifacts caused by metal implants in the past years. Iterative and interpolation are the two types of main. The iterative method is the commonly used method in Matrix Equation, which algorithm idea as follows:At first assumed that the image is uniform, compared the value of the theoretical calculated and the measured, and then modified the difference between the calculated and the measured, repeated until the difference is zero or within the permissible error range. For iterative correction method, the problem is how to accelerate the convergence rate and reduce computing time. The interpolation method is replaced projection of the metal implants with a synthetic projection generated by the adjacent sampling. These methods are relatively effective in overcome artifacts around the metal. However, any information on the metal object itself is completely lost. In addition, close to the area of metal objects is destroyed, and will bring new artifacts. I have summarized four methods to remove CT metal artifacts, there are based on the interpolation method, based on the normalization, based on wavelet multiresolution combined with linear interpolation and based on Frequency split method respectively.Firstly, several mathematical basis are described for understand the CT metal artifacts algorithm easily. There are threshold method, K-means cluster, total variation adaptive filtering and low-pass filter respectively. Threshold method is a regional Segment calculation, and particularly useful in gray-scale images which the object and the background occupy a different gray-scale range, small amount of calculation and stable performance are which point. In order to get the prior image model, K-means cluster algorithm is used, it is a very common dynamic clustering method and is a basic division method of clustering methods. Error square and guidelines function are often used as a clustering criterion function. The CT value of air, soft tissue, Bone and the metal region are reset for image clustering. Total variation adaptive filtering is used to reduce the streak artifacts in the original raw data before image reconstruction in Frequency split method. Which can adaptive control smoothing strength according to each projection of the gradient information in original raw data, so as to achieve better denoising effect and edge-preserving, Gaussian low pass filter is selected in the image frequency split processing, it is a smooth linear filter and can divide the image into the high-frequency component and the low-frequency component effectively.Secondly, the practice of the interpolation, the normalized and the wavelet multiresolution combined with linear interpolation method are introduced.And the advantages and disadvantages of the three algorithms are summarized. The linear interpolation algorithm is introduced, which idea is as follow:Segment the metal image in original metal artifact image, get the metal projection and make sure the interpolation area determined by the metal projection. Do an interpolation to metal projection by used nonmetal projection and plus it, reconstruct the correction projection. Although Interpolation method is relatively simple and largely to eliminate the streak artifacts and black shadows caused by metal implants. Which also bring two issues:New artifacts appear and any information on the metal object itself is completely lost, adjacent to the region of the metal objects are also destroyed. However, normalized method can solve the first issue. Which idea is as follow:do adaptive filtering to the original metal artifacts image so that reduce the streak artifacts and do k-means clustering in order to get the priori image model. And then do a projection for priori image model; do a projection for original metal image artifacts and dividing priori image projection to get the normalized projection; Segment the metal image in original metal artifact image and get the metal projection. Make sure the interpolation area determined by the metal projection; denormalized by multiplying the priori image projection after do a linear interpolation for the normalized projection, and then reconstructed the correction projection. Normalized method is the perfect solution to the first issue, in order to find a solution to the second issue, wavelet multi-resolution was considered. Because the main factors of the artifacts are low-frequency hardening and high-frequency noise. The useful signal mainly in the middle of frequency, the fine scales highlight the high frequency noise information highlights, the large-scale highlight the low frequency of hardening, the intermediate scale represent the useful information. So be imposed for these types of wavelet coefficients different weighting coefficients, protruding or weakening, So as to achieve the recovery of useful information but also the effect of suppressing artifacts. Which algorithm process is as follow:Segment the metal image in original metal artifact image and get the metal projection. Make sure the interpolation area determined by the metal projection; Do an interpolation to metal projection by used nonmetal projection and plus it. Make a wavelet transform for the original metal artifacts images and image after interpolation respectively. And do a weighted summation to the metal artifact image wavelet coefficients and the wavelet coefficients after interpolation. And finally get the corrected image. The results show that it can recover some of the useful information hidden in the destruction of data. But some bones information miss. It can be said that the three algorithms above have achieved a certain effect theoretically. But can’t be achieved relatively good results in actual project. For many clinical applications, the surgeons are very interested in the interface between the implant and adjacent bone and soft tissue. Therefore, an effective metal objects artifact correction method is hoped to retain more details near the metal region.removal of the metal artifacts completely. And no new artifacts appear.Finally, the frequency split remove metal artifacts method is introduced in details, which is the result of continuous practice, and which algorithm idea is as follow： First, Preprocess the raw data by the total variation adaptive filtering before reconstruct image.and then reconstruct the metal artifacts image; Segment the metal image in original metal artifact image and get the metal projection. Make sure the interpolation area determined by the metal projection; Denormalized by multiplying the priori image projection after do a linear interpolation for the normalized projection. And then reconstructed the correction projection; Do a Gaussian low-pass filtering for original metal artifact image and the normalized correction image for get low-frequency component. And get the high-frequency component by original metal artifact image and the normalized correction image subtracting the respective low-frequency component; Finally, Do a low-pass filtered for the metal image and normalized it get the weighted matrix. At last, weight the high frequency information of the original metal artifacts image, the high-frequency information and low-frequency information of normalization correction image to obtain the final corrected image. The frequency split solve the first issue by combine with normalization method, solve the second issue by reduce the steak artifacts by total variation adaptive filtering before reconstructed image and weighted the high frequency components of the reconstructed image.which restored edge information and structural details in the final corrected image. Removed the metal artifact completely and reserved more details near metal region in the meantime. Real patient data verified the validity and reliability of the algorithm. The frequency split method has great practical value for clinical diagnostic applications. It is truly a multi-ingredient, simple, effective, and easy to implement metal artifact correction method in practical engineering.