目的：探讨利用图像纹理特征对低剂量数字胸片不同拍摄条件进行定量分析的方法。方法分别对同一体模用12种不同拍摄条件拍摄3次数字X线摄影，并对图像进行基于灰度统计、基于灰度共生矩阵、基于小波的熵纹理特征提取，分别用欧几里得距离、余弦相似度和相关系数描述结节纹理特征与背景纹理特征的相似程度，并根据这些相似度指标确定最佳拍摄条件，最后与临床医生对拍摄条件的主观判断进行比较。结果分别提取了所有结节区域和背景区域的16个纹理特征，欧氏里得距离、余弦相似度和皮尔逊相关系数一致判定81 kV/1.0 mAs和81 kV/1.6 mAs的拍摄条件最佳，并与医生评价结果基本一致。结论通过特征相似度计算的方法，能够选中最佳拍摄条件下的图像质量，从而协助医生进行结节的分析，对疾病进行诊断。%Objective To explore the quantitative method of analyzing low dose digital radiography by using the image texture features.Methods The texture analysis was performed on 12 different shooting conditions of the same model. For each condition, 3 digital X-ray photography images were taken. The texture feature extractions of the images were based on the gray level statistics, the gray level co-occurrence matrix and the wavelet entropy. Euclidean distance, cosine similarity and correlation coefficient were used respectively to describe the texture feature similarity between the nodules and the backgrounds. According to the similarity indexes, the best shooting condition was determined and consistent with the subjective judgment on shooting conditions of clinical physicians.Results 16 texture features of all nodule regions and background regions were extracted respectively. Euclidean distance, cosine similarity, and Pearson correlation coefifcient consistently demonstrated that 81 kV/1.0 mAs and 81 kV/1.6 mAs were the best shooting conditions, and the evaluation results are basically identical with the doctors’ judgments.Conclusion By using the method of feature similarity calculation, we can select images with the best quality so as to help physicians to analyze the nodules as well as to observe and diagnose the disease.
展开▼
