Alternative Title医用画像における画質改善と評価手法の検討 -3D低線量MDCT画像の画質改善と定量評価-
Note (General)Advance in medical imaging technologies enables various imaging modalities such as ultrasound, MRI (magnetic resonance imaging), PET (positron emission tomography) and CT (computed tomography) to be used in medical diagnosis. However, their imaging performance is limited by various constraints. That causes degradation of the medical image quality. Especially, the hazard associated with X-ray exposure is greater for 3D-CT methods such as MDCT (multi-detector row computed tomography) than for conventional CT. The best method of mitigating this hazard is to lower the X-ray dose level. However, the radiographic noise, i.e., quantum noise, in MDCT images increases when radiation exposure is reduced. Therefore, there is a strong desire to reduce patient dose and to improve image quality by increasing spatial resolution and decreasing quantum noise in MDCT modality. Although various image improvement and evaluation methods have been developed, they all focused on improving the image quality in a whole image or in an entire 3D dataset. Their performance is not enough for the radiologists who should not miss the small tumors whose size is around 5 mm and preferably down to 2 mm. To overcome these difficulties, this thesis proposes several linear and non-linear filtering methods to enhance the low-dose MDCT images according to the size and local voxel value distribution of the small regions which are suspected of tumors. Firstly, 3D weighted average filter is designed to enhance the low-dose MDCT images. In order to retain the subtle characteristics of the small tumors, the center weight of the 3D weighted average filter is given a large value. The other weights decrease with the increasing of the radius away from the center of the filter template. However, the 3D weighted average filter is a linear filter. Therefore, edges and subtle structure of parenchyma are prone to be blurred to some degree after filtering. To alleviate this limitation, 3D non-linear diffusion-based filter is designed to reduce the quantum noise as well as to maintain the subtle structure of the small tumors and the parenchyma. Experimental results show that the edges of the tumors remain well while the noise level has been lowered by this diffusion type filter. However, the point noises of the images are not be removed well, since this diffusion type filtering method can not distinguish noise candidates from the voxels with large gradients. Therefore a combination filter consisting of 3D weighted average filter and 3D non-linear diffusion based filter are proposed. To improve the quality of low-dose clinical MDCT images degraded by the quantum noise, a new 3D adaptive median filter with local average (3D AMLA) is also proposed. This new method differs from approaches in the literature since not only it is designed based on the size of the small tumors but also this scheme detects and replaces the noise candidates with local averaging according to the distribution of voxel value adaptively. In this study, the 2D adaptive median filter (AM) is extended to 3D AM to show the effectiveness of 3D filtering, firstly. Then adaptive local averaging strategy is introduced to this 3D AM. That is 3D AMLA. The 2D AM, 3D AM and 3D AMLA are applied to 80-, 60-, 40- and 20%-dose low-dose datasets. Image quality is assessed by visual evaluation, voxel value profiles and several quantitative methods, including newly proposed entropy of correlation (EC) and mutual information and correlation (MIC) which are developed according to the graininess correlation of the little intro-hepatic tumors in directions of X, Y and Z axes. Experimental results show the high performance of both the filtering and the evaluating methods and X-ray dose level can be reduced to a large degree by applying the 3D AMLA and the combination filter.
Collection (particular)国立国会図書館デジタルコレクション > デジタル化資料 > 博士論文
Date Accepted (W3CDTF)2020-01-16T18:57:33+09:00
Data Provider (Database)国立国会図書館 : 国立国会図書館デジタルコレクション