基于小波变换的超声图像纹理特征提取及前列腺癌诊断

Diagnosis of Prostate Cancer and Texture Feature Extraction of Ultrasound Images Based on Wavelet Transform

目的根据小波变换原理,研究了前列腺直肠超声图像中纹理特征的提取方法,并应用于前列腺癌的早期诊断。方法本文提取出前列腺直肠超声图像中目标区域的小波变换纹理特征和边界频率特征,通过主分量分析方法(principal components analysis,PCA)对提取出的纹理特征进行选择,得到一个最优的特征子集。然后分别应用K均值聚类、支持向量机(support vector machine,SVM)算法和AdaBoost(a-daptive boosting)算法来对所提取出的病变区域纹理特征进行分类。结果对比实验结果表明,本文所提取的特征比单纯的使用灰度级差矢量(gray level difference vector,GLDV)具有更好的区分良恶性图像的能力,AdaBoost算法和SVM算法都能够有效地识别病变区域,识别正确率达到94.12%和93.46%。结论使用本文算法可以为医生诊断提供有用的辅助信息,并提高诊断效率。

Objective To study the texture feature extraction of prostate ultrasound images based on the wavelet transform for the early diagnosis of prostate cancer. Methods This paper extracted the wavelet texture features and edge-frequency features from pathological regions in transrectal ultrasound images,then the reduced optimal feature set was selected by principal components analysis（PCA） algorithm,and the classification was done by K-means,support vector machine（SVM） and AdaBoost algorithm individually. Results We compared the texture features with Mohamed＇s,the experiment results showed that the extracted features had the better ability to differentiate the benign or malignant images than the mere gray level difference vector （GLDV）. AdaBoost and SVM could differentiate the pathology regions efficiently and gave the identify rate of 94.12%,93.46% respectively. Conclusion The proposed algorithm can supply useful information to the doctors for the clinical diagnosis and the diagnosis efficiency is enhanced.