Research On Metalartifact Reduction In CT Medical Image

Computed tomography has been one of the most important diagnostic method inclinical medicine nowadays. However, the metal artifacts in CT images influence thediagnosis. Therefore, the metal artifacts reduction is one of challenging key issues.We analyze why the metal artifacts are formed in CT images. Because of the higherdensity, a metal object attenuates the diagnostic x-rays more greatly than soft tissues andbone. Much fewer photons can be received by detectors, so the received data are withsome mistakes.. After the filtered back-projection (FBP), the metal will be present in thereconstruction image, and induces severe dark and bright streaks. These artifactsseriously degrade image quality particularly near the metal surfaces; therefore they willmake diagnosis very difficult or even impossible.In this thesis, our work focus on the metal artifacts reduction. By improving thetypical MAR(Metal Artifact Reduction) algorithms and NMAR method respectively, wepropose two metal artifacts reduction methods. The details are as follows.1. We propose an improved MAR scheme based Cosine projection integration.On the typical MAR1framework, we predict the metal trace from the original sonogramdirectly. Our algorithm do not need to reconstruct the original images and it is timesaving. Furthermore, even if there is large metal artifacts our algorithm could get thestable results. The experimental results confirm the efficiency of the proposedalgorithm.2. We propose a Trace-based NMAR algorithm. Based on the framework ofNMAR algorithm, we introduce a cosine trace-based repairing algorithm to compensatethe missed data on the cosine trace. It improves the discontinuity of datas along thetrace direction. Experiments show that the proposed algorithm obtains good results,especially for the artifacts between the metal and tissues of high density.3. We introduce bilateral filter to TNMAR, so that the metal area could besegmented effectively. This algorithm is the trade off between spatial information andgray level similarity. It can simultaneously reduce the noise and preserve the edgeinformation in the image. The experimental results show that it’s suitable for reducingthe noise in the CT images.