Advanced Noise Reduction Processing for X-ray CT System with Iterative Processing. Koichi Hirokawa MEDIX VOL. 56 P.43 P.46

Transcription

1 Advanced Noise Reduction Processing for X-ray CT System with Iterative Processing Taiga Goto Koichi Hirokawa Hisashi Takahashi MEDIX VOL. 56 P.43 P.46

2 Advanced Noise Reduction Processing for X-ray CT System with Iterative Processing Taiga Goto Koichi Hirokawa Hisashi Takahashi CT Systems Division, Hitachi Medical Corporation Our primary goal is to develop a genuine iterative reconstruction method as a next-generation high-quality imaging technique. However, the routine use of such a technique for clinical examinations is currently difficult owing to calculation costs that are several tens to hundreds of times higher than those of conventional reconstruction procedures. Therefore, we developed an iterative noise reduction technique (Intelli IP 1 )that was subsequently implemented in SCENARIA 2. In this paper, we introduce a second more advanced version of Intelli IP, as a technique that achieves higher noise reduction. Key Words: Iterative Processing, Noise Reduction, Intelli IP 1. Introduction X-ray computed tomography (CT)has become firmly established as a diagnostic apparatus that provides highdefinition morphology information based on the multislice technique, improved high spatial resolution and high temporal resolution due to high-speed rotation and ECGgated scanning. CT is in high demand because of its high diagnostic value and simplicity of use in the examination. It was in this context that Gonzalez reported the effects of X-ray dose in ), which garnered increased interest in the effects of radiation dose and increased importance placed on reducing the exposure dose. However, merely reducing the radiation dose during scanning increases noise in the reconstructed image and degrades performance characteristics such as lesion detectability. Increased noise associated with future improvements in spatial resolution is also a concern, and therefore the development of noise reduction techniques is very important from this standpoint as well. We have therefore taken approaches to reduce noise from both hardware and software perspectives, such as by enhancing detection efficiency, reducing DAS noise, improving reconstruction filters, and developing AEC techniques (IntelliEC 3 )as well as nonlinear image filter techniques (adfilter 4 ) 2). Currently, we are continuing to develop an iterative noise reduction technique that takes statistical models into account. The results to date of this work have been implemented in SCENARIA 2 as Intelli IP 1 (Normal) *1 and Intelli IP (Advanced). 2. Iterative noise reduction processing (Intelli IP)considering statistical models and anatomical structure 2.1 Noise in the projection space and streak artifacts in the image space Detected X-ray photons exhibit statistical fluctuations which are known to follow a Poisson distribution. Projection data include statistical photon fluctuations (photon noise)and electrical noise from the scanner system. This means that the higher projection data (i.e., projection data from photons through larger object)include a higher amount of noise from a greater shortage of photons. In CT, photon noise in the projection data appears as a streak artifact in the reconstructed image (Fig. 1). For this reason, cases in which the number of photons is insufficient along a particular direction, such as the longitudinal direction of the shoulders, show streak artifacts that intensify along this direction and which are more likely to become evident in the reconstructed image. MEDIX VOL.56

3 Here, if image noise is fine-grained, it can be reduced rather easily by noise reduction processing on the reconstructed image. In contrast, streak artifacts are difficult to distinguish from structural objects in the reconstructed image. Thus, the reduction of streak artifacts on the reconstructed image can cause intense blurring of structural objects. For this reason, the reduction process of streak artifacts must be performed in the projection space. Hence, it is crucial to improve the precision of noise reduction processing in the projection space. 2.2 Roadmap for developing a noise reduction technique based on iterative processing We are developing a noise reduction technique based on iterative processing that considers statistical models. Ultimately, as a genuine iterative reconstruction method, we Noise Projection space Reconstruction Streak artifact Image space Fig. 1: Noise in the projection space and streak artifact are aiming for a maximum noise reduction of approximately 60% *2 using algorithms containing characteristics that are unique to our company 3). At the same time, we have made stepwise advances in the development of Intelli IP as an iterative noise reduction processing that also places emphasis on rapid processing speed. Until now, we have equipped SCENARIA with Intelli IP (Normal), which emphasizes processing speed in particular. For the Intelli IP (Advanced)implemented here in SCENARIA, we have improved its ability to reduce radiation dose at practical processing speeds (Fig. 2). 2.3 Difference between Intelli IP (Normal)and Intelli IP (Advanced) (1)Intelli IP (Normal) The Intelli IP (Normal)currently implemented in SCE- NARIA performs noise reduction on both projection and image data by adaptive iterative processing based on statistical reliability. This processing technique optimizes properties such as the degree of noise reduction, degree of sharpness and graininess for each anatomical region. Specifically, noise reduction processing in the projection space is performed selectively and iteratively for low-reliability projection data based on statistical reliability (Fig. 3). In the image space, iterative processing that selectively reduces the noise is performed in accordance with several types (starting with A )of optimized parameters for each anatomical region. Processing speed is particularly Intelli IPNormal : Installed on SCENARIAImage noise reduction: Up to 30% 2 Adaptive iterative processing based on statistical reliability Processing speed is emphasized for real-time processing Fig. 2: Noise reduction technique applying iterative processing Intelli IPAdvanced : Newly installed on SCENARIAImage noise reduction: Up to 40% 2 Reconstruction applied with iterative processing using a more strict statistical model The balance between processing speed and image quality is emphasized. Iterative reconstruction : Under development for future CT models Image noise reduction: Up to 60% 2 Genuine iterative reconstruction based on a statistical model High image quality is pursuedseveral tens to hundreds of times higher than those of conventional reconstruction procedures Advanced algorithm with unique characteristics compared with the existing iterative reconstruction methods 11th International Meeting on Fully Three-Dimensional Image Reconstruction in Radiology and Nuclear Medicine, July 11 - July 15, 2011, Potsdam, Germany 67th Annual Meeting of Japanese Society of Radiological Technology High image quality High speed Projection space Noise reduction processing based on a statistical mode Intelli IP Back projection Image space Image quality tune based on an anatomical model Iteration Iteration Fig. 3: Overview of Intelli IP processing MEDIX VOL.56

4 emphasized in Intelli IP Normal, in which real-time pro- placed on the balance between processing speed and image cessing has been realized and image noise reduction up to quality. As a result, an image noise reduction effect of up to approximately 30 2 has been achieved by simplifying approximately 40 is achieved along with a streak artifact the statistical processing. reduction effect Figs. 4 6 in the next section. Currently, 2 Intelli IP image reconstruction using Intelli IPcan be The Intelli IPimplemented here in SCE- performed in a practical processing time that is about two NARIA is an iterative processing in which the statistical to three times longer than those without using iterative processing precision has been improved as compared with processing. However, increased computer processing Intelli IP Normal. Specifically, Intelli IPis a resources may be employed to reduce the time necessary processing that, after reducing the noise in the projection for the increased reconstruction workload. space based on a high-precision statistical model by an iterative processing, tunes image quality based on anatomical and statistical information in the image space. Hence, it is able to greatly reduce the amount of image noise and streak artifacts. With Intelli IP, emphasis is SD=24.04 Fig. 4: Reconstructed Images Mediastinum, WW: 300, WL: Performance characteristics of Intelli IP Effects of noise reduction 1 Reconstructed images Fig. 4 to Fig. 6 show the results of applying the conven- SD=16.96 a Conventional reconstruction b Intelli IP Normal, E SD=69.51 SD=47.08 Fig. 5: Reconstructed Images Lung field, a Conventional reconstruction b Intelli IP Normal, E WW: 1000, WL: 650 SD=26.72 Fig. 6: Reconstructed Images Abdomen, WW: 200, WL: 60 SD=23.08 a Conventional reconstruction b Intelli IP Normal, E SD=14.27 c Intelli IP SD=42.35 c Intelli IP SD=15.88 c Intelli IP MEDIX VOL.56

5 tional reconstruction method (Intelli IP = OFF), Intelli IP (Normal, E) *1, and Intelli IP (Advanced), respectively. In the results here, by measuring the image noise (image SD)with the ROI positions in the figures, noise reduction effects of 10-30% with Intelli IP (Normal, E)and approximately 40% with Intelli IP (Advanced)relative to the conventional reconstruction method (Intelli IP = OFF) are obtained (Table 1). In addition, from the same figures, it can be seen that streak artifacts are reduced more effectively than Intelli IP (Normal, E), because the statistical model is more properly considered in Intelli IP (Advanced). At the same time, there is very little blurring of structural objects such as those present in bone tissue. (2)Spatial resolution (MTF: Modulation Transfer Function) Fig. 7 shows the measurement results of MTF (Modulation Transfer Function)measurements about the images using the standard reconstruction filter for abdomen. As is evident from the figure, no degradation of the MTF occurred compared to the conventional reconstruction method (Intelli IP = OFF)for any of types A-E with Intelli IP (Normal). Then with Intelli IP (Advanced), there is substantially no degradation about the frequency values at which MTF values show 10% and 50%. 3. Conclusion We have implemented Intelli IP (Advanced)in SCE- NARIA by improving Intelli IP (Normal). Intelli IP (Advanced)can, for example, reduce streak artifacts arising from local photon shortages, and attain image noise reduction up to approximately 40% by performing noise reduction with an iterative processing that considers statistical models more properly. We therefore consider this to be a promising technique for improving the image quality results, when considering the trade-off between image quality and radiation dose. However, given the importance of reducing radiation dose as much as possible, we will continue to aim for the commercialization of a next generation genuine iterative reconstruction method and advance its development in the future. 1 Intelli IP, 2 SCENARIA, 3 IntelliEC, and 4 adfilter are registered trademarks of Hitachi Medical Corporation in Japan and/or other countries. *1 For convenience of distinction, we refer to the conventional Intelli IP as Intelli IP (Normal), and cases in which type E is used in Intelli IP (Normal)as Intelli IP (Normal, E). *2 Results when compared to conventional reconstruction (Intelli IP=OFF). Table 1: Image noise Image SDmeasurement results Mediastinum Lung Field Conventional IPNormal, E IPAdvanced Abdomen MTF % reduced % reduced % reduced % reduced % reduced % reduced Conventional IPNormal, A IPNormal, B IPNormal, C IPNormal, D IPNormal, E IPAdvanced References 1) Gonzalez, et al.: Risk of cancer from diagnostic X-rays: estimates for the UK and 14 other countries. The Lancet, Vol.363: , ) T. Nakazawa, et al.: The new technologies of the multislice CT system SCENARIA, MEDIX, Vol.55: 45-48, ) H. Takahashi, et al.: Motion Tolerant Iterative Reconstruction Algorithm for Cone-Beam Helical CT Imaging, 11th International Meeting on Fully Three- Dimensional Image Reconstruction in Radiology and Nuclear medicine: , Spatial frequency [lp/cm] Fig. 7: MTF results standard reconstruction filter for abdomen MEDIX VOL.56

6 International Operations Group Akihabara UDX, , Soto-Kanda, Chiyoda-ku, Tokyo, Japan Phone: Telefax: URL XX-E131