腹部主动脉瘤无增强CT管腔识别的后处理算法研究
本研究评估一种数学算法在无增强CT上识别腹主动脉瘤管腔的可行性。对233张CT图像提取管腔并与增强CT金标准比较,像素误分类误差En均值0.14,面积差异ΔA均值0.28,提示算法可较准确检测管腔位置并提取轮廓。
要点
- 无增强CT经Kantorovich增强与小波分解可提取腹主动脉瘤管腔轮廓。
- 233张图像中像素误分类误差En均值0.14,面积差异ΔA均值0.28。
- 结果提示未来或可减少CT对比剂使用,尚需进一步临床验证。
核心结论
- 所开发的数学算法能较准确检测血管管腔位置并提取其轮廓。
- 研究结果令人鼓舞,提示未来可能用于减少CT对比剂使用。
研究设计与样本
- 对腹主动脉瘤患者的增强与无增强CT进行对比评估。
- 算法首先手动勾画目标主动脉区域附近的方形感兴趣区(ROI),其余流程自动完成。
- 算法流程包括Kantorovich算法增强、小波分解、低频成分选择及基于灰度直方图的阈值分割。
- 以增强CT扫描作为金标准,共应用于233张CT图像。
主要发现
- 数值误差En均值范围为0.12至0.17,面积差异ΔA均值范围为0.11至0.35。
- 全部233张CT图像序列中,En均值0.14,ΔA均值0.28。
临床怎么用
- 该算法可在无增强CT上辅助定位腹主动脉瘤管腔并提取轮廓,为减少对比剂使用提供潜在途径。
以上解读依据论文摘要(全文未开放获取或许可不允许翻译转载)。
原文
INTRODUCTION: Contrast medium (CM) use in computed tomography (CT) is limited by nephrotoxicity and possible allergic reactions. The purpose of this study is to introduce a tool for the diagnosis of abdominal aortic aneurysms (AAAs) by avoiding the use of CM. METHODS: With and without CM CTs of patients with AAA were evaluated. A mathematical algorithm was implemented to allow visualization of the inner aortic lumen in the series without CM. The first step of the algorithm consisted in manually highlighting a squared region of interest (ROI) close to the target aortic area. The rest of the algorithm is completely automated and performs the following flow of operations: The "Kantorovich" algorithm is applied to the ROI for image enhancement. Then, a wavelet decomposition method is applied to identify the different frequency components of the image. Exploiting the wavelet decomposition, the system selects the low-frequency components of the image, corresponding to the major structures. Thresholding method, based on the analysis of the gray-level histogram, is then performed to extract the contours of the vessel. At this point, the extraction of the pervious area is completed. Final images were compared with the contrast enhanced scans, valued as gold standard. To validate the algorithm, an analysis of the results has been performed considering the following types of error: En = #m/#ROI ΔA = |1-(#CM/#CEX)| (#m = number of misclassified pixels; #ROI = number of pixels in the ROI; #CM = number of pixels belonging to the CM; #CEX = extracted areas). En provided a measure on the pixels wrongly classified, and ΔA the difference of extracted areas. RESULTS: The algorithm was applied to 233 CT images. Extracted images were compared with the ones with CM. Mean values of the numerical errors ranged from 0.12 to 0.17 for En and 0.11 to 0.35 for ΔA. For all the 233 CT images in the analyzed sequences, the mean error was 0.14 (En) and 0.28 (ΔA). CONCLUSIONS: The developed mathematical algorithm allows to detect the position of the vessel lumen and to extract its contours with a good accuracy. Our experience shows encouraging results and suggests a possible future clinical application to reduce CT CM use.
引用
- Cieri E, Simonte G, Costarelli D, et al. Computed Tomography Postprocessing for Abdominal Aortic Aneurysm Lumen Recognition in Unenhanced Examinations. Ann Vasc Surg. 2019. PMID: 31200051. doi:10.1016/j.avsg.2019.05.002
供执业医师工作参考,不替代临床判断;数值与分级以指南原文为准。