CT画像における早期虚血サイン検出アルゴリズムの開発

抄録

This paper describes an approach to automatically detect the parenchymal hypoattenuation of hyperacute stroke in nonenhanced computed tomography (CT) images. This technique is based on a comparison between the CT values histograms of right and left brains. A subtraction curve that was regarded as an output value was calculated by subtracting the right-hemispherical histogram from the left-hemispherical histogram obtained from one of the region-of-interest (ROI) sets on an image. The output value was used to assess whether hypoattenuation exists on CT images with a threshold value. If judged abnormal, a rectangular region including a whole or partial hypoattenuation area was detected from the ROI. Twenty-six cases with hypoattenuation and 30 cases without hypoattenuation were included in this study. As a result of our experiments, the sensitivity of this method in detecting hypoattenuation was found to be 92%, with approximately 0.16 false-positive per image. Our preliminary experimental results indicated that the proposed technique can be used for the automated detection of parenchymal hypoattenuation of hyperacute stroke on nonenhanced CT images.