@phdthesis{oai:kitami-it.repo.nii.ac.jp:00007858,
 author = {DONG, JIAN and 董, 建},
 month = {Sep},
 note = {Chapter VIII Research summary and conclusion

When metallic prosthetic appliances and dental fillings exist in the oral cavity, the appearance of metal-induced streak artifacts is not avoidable in CT images. The general objective of this research work was to reduce the metal-induced streak artifacts using successive iterative reconstruction method. ML-EM and OS-EM algorithms were analyzed and used in this study. Besides streak artifact reduction, 3D filtering method and region growing method were applied to improve image quality., 8.1 Research summary

In Chapter I, discipline of medical imaging was introduced. Among several medical imaging modalities, X-ray CT was emphatically introduced from its tomography principles to its application.  Artifacts which were usually observed on CT images were analyzed and sorted to several kinds. Although the traditional CT reconstruction method, the FBP algorithm, had been widely applied, it could not deal with the loss of portions on projection data which was caused by extremely high X-ray absorption coefficients of metallic biomaterials. Image reconstruction algorithms that can use the corrupted projection data were expected. 
In Chapter II, the regularly used image format, DICOM, was introduced firstly.It is the abbreviation of digital imaging and communication in medicine. Fundamental composition of DICOM images was expressed. Special information about patient and examination are contained in DICOM images. Image size, bits allocation were analyzed subsequently. Two medical image displaying software, Image J and OsiriX, were introduced. We used Image J to display and analyze 2D images. We used OsiriX to present 3D volume rendering.
In Chapter III, projection data acquisition on X-ray CT modality was instructed in detail. All the projection data were acquired in 360 directions with 1 degree intervals in this study. Then, iterative restoration methods as CT image reconstruction algorithms, ML-EM and OS-EM, were proposed. The two algorithms both result in an approximation between the processing image and the
target image. A 3 × 3 matrix reconstruction by ML-EM algorithm was simulated. It simply demonstrated the image reconstruction process. Only 26 iterative calculations reached the target matrix. However, much more iterations were estimated to be needed for reconstructing 512 × 512 CT images. Subsequently, convergence validation of both ML-EM and OS-EM algorithms were carried out on practical images. It can be concluded that almost 300 iterations were needed to reconstruct the practical image by ML-EM, while the iteration times was about 50 for OS-EM algorithm to reconstruct the same image. The OS-EM algorithm can reconstruct image data faster without image quality dropped and it is usually used to reduce the calculation time.
In Chapter IV, variable optimization of ML-EM algorithm was tested when using in streak artifact reduction. 50 iterations could reach a best artifact reduction effect. More than 50 iterations could not improve the effect any more. Then we applied the ML-EM on a sequential CT images to reduce artifacts using the same artifact-free image’s projection data. For OS-EM algorithm, the best parameter combination was proved to be subset = 8 and iteration = 10.
In Chapter V, we proposed the successive iterative reconstruction method (SIRM) based on the fact that adjacent CT images often depict very similar anatomical structures within the resulting collection of thin-slice images. We used SIRM in streak artifact reduction in dento-alveolar region. First the projection data of the artifact-free slice was obtained. The adjacent slice, which showed weak artifacts, was processed using the artifact-free slice’s projection data. Then the projection data of the newly processed image was obtained and it was used to reconstruct the next image which contained a little more artifacts. In this manner, the processing continued until the heave streak artifact containing image was processed. Two alternative algorithms were tested in processing sequential images for artifact reduction with SIRM.
In Chapter VI, SIRM was clinically applied in streak artifact reduction. Images before and after sagittal split ramus osteotomy operation (a jaw deformity case) were processed. The method was also proved to be effective in reducing artifacts caused by orthodontic wire, brackets, bone screws or titanium plates. Calculationacceleration was realized by general purpose graphic processing unit (GPGPU).
In Chapter VII, SIRM was applied in cone-beam CT images for streak artifact reduction. Cone-beam CT has become increasingly important in treatment planning and diagnosis in dentistry of endodontics or orthodontics. Besides, 3D filtering method and region growing method with dilation and erosion were employed for image quality improvement., 8.2 Conclusion

The dissertation has presented a better solution for metal-induced streak artifacts reduction in X-ray CT images. ML-EM and OS-EM algorithm were both proved to be effective in artifacts reduction. Successive iterative reconstruction method reduced the deviations on edge of anatomical structure. GPGPU processing realized the calculation acceleration. Image quality improvement was also realized in three dimensional level.},
 school = {北見工業大学},
 title = {Three-Dimensional Image Processing for Artifact Reduction and Quality Improvement in Medical X-ray Computed Tomography},
 year = {2014}
}