检索历史 清除

目的:提出一种基于Canny边缘检测算法求取CT胸腔图像噪声的方法。方法:选取2021年采集的男性志愿者的CT胸腔图像250张，通过Canny自适应阈值法提取CT胸腔图像轮廓，并比较Sobel算法、Canny双阈值法和Canny自适应阈值法的相似度；使用霍夫变换来确定感兴趣区域，考察感兴趣区域的选取大小、重建卷积核和管电流对CT胸腔图像噪声的影响。结果:Canny自适应阈值法保留了更多细节，边缘的连续性和完整性有所提高，对边缘检测与图像分割更加灵活以及更加具有鲁棒性。Canny自适应阈值法结构相似性指数最高（0.644），均方根误差最小（0.371），其在边缘轮廓检测方面相似度最高，效果更显著。随着方形感兴趣区域大小的增加，平均噪声呈现下降趋势，噪声标准差在某些区间有所增加，特别是在较大的方形区域。在同一重建卷积核的情况下，CT胸腔图像升主动脉的平均噪声比胸主动脉的高，升主动脉的噪声标准差比胸主动脉的低。对于升主动脉，重建卷积核E的升主动脉平均噪声（41.97 dB）最低，噪声标准差（20.64 dB）最大；对于胸主动脉，重建卷积核E的平均噪声（30.78 dB）最低。胸主动脉的平均噪声和噪声标准差随管电流的增加而下降。结论:提出了一种基于Canny边缘检测算法求取CT胸腔图像噪声的方法，适用于检测CT胸腔图像。

Objective:To propose a method for obtaining noise from thoracic CT images based on the Canny edge detection algorithm.Methods:A total of 250 pieces of thoracic CT images of male volunteers collected in 2021 were selected. The contours of thoracic CT images were extracted by the Canny adaptive threshold method. The similarity of the Sobel algorithm, the Canny double threshold method, and the Canny adaptive threshold method was compared. The Hoff transform was used to determine the regions of interest. The effects of the selection size of the regions of interest, the reconstructed convolution kernel, and tube current on the noise of thoracic CT images were investigated.Results:The Canny adaptive threshold method preserved more detail, and the continuity and integrity of the edges were improved, indicating that it was more flexible and robust for edge detection and image segmentation. The Canny adaptive threshold method had the highest structural similarity index (0.644) and the lowest root mean square error (0.371). It had the highest similarity in edge contour detection, and the effect was more significant. As the size of the square area of interest increased, the average noise decreased. The noise standard deviation increased in some intervals, especially in larger square regions. In the case of the same reconstructed convolution kernel, the mean noise of the ascending aorta in thoracic CT images was higher than that of the thoracic aorta. The noise standard deviation of the ascending aorta was lower than that of the thoracic aorta. For ascending aorta, the mean ascending aorta noise (41.97 dB) of reconstructed convolutional nucleus E was the lowest, with the highest noise standard deviation (20.64 dB). For the thoracic aorta, the mean noise (30.78 dB) of the reconstructed convolutional nucleus E was the lowest. The mean noise and standard deviation of the thoracic aorta decreased with the increase in tube current.Conclusions:A method based on the Canny edge detection algorithm to obtain noise from thoracic CT images is proposed, which is suitable for detecting thoracic CT images.