中国医药指南
中國醫藥指南
중국의약지남
Guide of China Medicine
2015年
25期
2-3
,共2页
肝脏%良恶性病变%灰度共生矩阵%纹理特征%决策树
肝髒%良噁性病變%灰度共生矩陣%紋理特徵%決策樹
간장%량악성병변%회도공생구진%문리특정%결책수
Liver%Benign and malignant lesions%Gray symbiotic matrix%Texture feature%Decision tree
目的：应用灰度共生矩阵和决策树分类的挖掘的方法对肝脏B超纹理特征进行分析，探讨肝脏B超影像纹理特征在肝脏恶性病灶中的应用。方法随即选取120例正常肝脏、肝脏良性病变，肝脏恶性肿瘤的肝脏B超影像进行增强去噪处理，通过构建反映共生矩阵各角度信息的灰度共生矩阵提取纹理特征参数，结合决策树算法进行分析诊断（所有患者术前均进行二维超声，术后经病理手术确认）。结果实验表明该方法对整个肝脏典型病理影像分类的准确度达到83.33%，在判断恶性病变时，查全率为83.3%，查准率为73.9%，调和均值F_mean 90.9%，接受者操作特征（ROC）85.3%，具有较高的诊断率。结论该方法是分析肝脏影像图的一种快速有效的纹理特征分析方法。提取的纹理特征对图像内容有较好的分区性，为临床上辅助诊断肝脏疾病提供了量化依据，也为后期图像识别，图像数据挖掘和图像检索提供了很好的特征数据。
目的：應用灰度共生矩陣和決策樹分類的挖掘的方法對肝髒B超紋理特徵進行分析，探討肝髒B超影像紋理特徵在肝髒噁性病竈中的應用。方法隨即選取120例正常肝髒、肝髒良性病變，肝髒噁性腫瘤的肝髒B超影像進行增彊去譟處理，通過構建反映共生矩陣各角度信息的灰度共生矩陣提取紋理特徵參數，結閤決策樹算法進行分析診斷（所有患者術前均進行二維超聲，術後經病理手術確認）。結果實驗錶明該方法對整箇肝髒典型病理影像分類的準確度達到83.33%，在判斷噁性病變時，查全率為83.3%，查準率為73.9%，調和均值F_mean 90.9%，接受者操作特徵（ROC）85.3%，具有較高的診斷率。結論該方法是分析肝髒影像圖的一種快速有效的紋理特徵分析方法。提取的紋理特徵對圖像內容有較好的分區性，為臨床上輔助診斷肝髒疾病提供瞭量化依據，也為後期圖像識彆，圖像數據挖掘和圖像檢索提供瞭很好的特徵數據。
목적：응용회도공생구진화결책수분류적알굴적방법대간장B초문리특정진행분석，탐토간장B초영상문리특정재간장악성병조중적응용。방법수즉선취120례정상간장、간장량성병변，간장악성종류적간장B초영상진행증강거조처리，통과구건반영공생구진각각도신식적회도공생구진제취문리특정삼수，결합결책수산법진행분석진단（소유환자술전균진행이유초성，술후경병리수술학인）。결과실험표명해방법대정개간장전형병리영상분류적준학도체도83.33%，재판단악성병변시，사전솔위83.3%，사준솔위73.9%，조화균치F_mean 90.9%，접수자조작특정（ROC）85.3%，구유교고적진단솔。결론해방법시분석간장영상도적일충쾌속유효적문리특정분석방법。제취적문리특정대도상내용유교호적분구성，위림상상보조진단간장질병제공료양화의거，야위후기도상식별，도상수거알굴화도상검색제공료흔호적특정수거。
Objective To study the application of the liver B ultrasonic image texture feature in malignant liver lesions through the method of data mining on liver B ultrasonic texture image feature analysis, based on gray level co-occurrence matrix (GLCM) and decision tree classification. Method 120cases of liver B ultrasound image of normal, benign and malignant tumors were selected for analysis. After enhancement denoising processing, the parameter information of texture feature was extracted through constructing the GLCM, reflecting angle information of co-occurrence matrix. Then analysis diagnosis was performed in combination with decision tree algorithm (all of the patients were examined with preoperative 2D ultrasound, and confirmed by the pathological examination).Results Using this method, the liver typical pathological image classification accuracy can reach 83.33%. For malignant lesions, the recall rate was 83.3%, the precision rate was 73.9%, the harmonic mean F_mean was 90.9% and receiver operating characteristic (ROC) 85.3%. These results show that this method has higher diagnostic rate.Conclusion Texture features calculation method in this paper is a rapid and effective method to analyze the liver B ultrasonic texture image feature, with higher classification accuracy than other methods. This method may be an effective way for clinical assistant diagnosis. It can provide quantitative basis for diagnosis of liver disease. Besides, it also provides typical data for image recognition, data mining and image indexing.
