Evaluation of Signal and Noise and Exposure in Medical Image for CAD

点击翻译按钮获取中文摘要

We have confirmed that the spatial resolution difference of CT images among CT scanners certainly influenced the diagnostic performance of CAD. To obtain the robustness of CAD for the dependency on CT scanners and improve the performance of CAD, we have developed the method to transform the acquired CT images to the standard ones by using the 3-D LSF which was measured for the considering CT scanner ; and, we have shown that the diagnostic performance of some CADs was improved by this method.Further, we have assessed the method of estimating image noise SD on CT images, and have shown that the method for estimating the noise variance of an image proposed by Rank et al. will be able to estimate the CT image noise SD with reasonable accuracy, unless the neighborhood pixels does not too much correlate with each other. We have also shown that streak artifacts on CT images can be statistically modeled by the Gumbel distribution, and, by using this result, we have devised the new evaluation method of streak artifacts on CT images that we call as the Gumbel evaluation method. Further more, we have assessed whether or not the fractal-feature distance could be a substitute for the observer performance index in Contrast-Detail (C-D) examination and image-reading study for evaluation of anatomic noise ; and, we have shown that all C-D and chest images can be analyzed by the virtual volume method and that the fractal-feature distances would be equivalent to IQFs and observer's confidence levels.In addition, we have measured the dose distributions of various organs and tissues during examinations with multi-detector row CT, under the irradiation conditions close to actual clinical examinations, and derived the corresponding effective doses. Further, we have evaluated the relationship between radiation dose and various image quality indices.