一种胸部X射线摄影图像中结节检测的多尺度匹配滤波器

摘要

为了提高胸部X射线摄影图像中肺结节的正确检测率,提出一种基于通道化Hotelling模型的多尺度匹配滤波算法.以X射线胸片中的正样本和负样本图像作为训练集合,用Hotelling分类器训练出不同尺度的匹配滤波器组,并与待检测的胸片图像做滤波.计算滤波输出图像的视觉检测差异,以最大差异为约束条件寻找最优尺度的滤波器,计算待检测图像中的结节尺寸和检测能力.实验结果表明,基于通道化Hotelling的匹配滤波算法的病灶检测能力与主观评价结果取得了较好的一致性,并具有较高的病灶检测准确率,滤波处理后的图像中仅含少量的假阳性区域,计算出的病灶的相对半径误差主要位于-0.21和0.18之间,该误差范围的图像占实验数据库图像的比例为86%.%A multi-scale matched filter based on Hotelling classifier is presented to improve the accuracy of nodule detection in chest X-ray radiography. The positive and negative radiography samples are used to train multi-dimension matched filters by Hotelling model. The visual detection difference is calculated using the filter outputs of radiography,and the optimal dimension filter is searched subject to the maximum detection difference. Then the nodule size and detection capability are calculated based on the optimal filter. Experimental results show that there is good accordance of the detection performances between the presented method and subjective observers.The method improves the detection accuracy and reduces the false-positive regions in the processed images. Errors of the calculated relative nodule sizes mainly distribute between -0. 21 and 0.18 and the proportion of the images in this range is 86％.