应用去噪算法的乳腺肿块超声辅助诊断研究

Computerized detection and classification on breast ultrasound using speckle reduction

目的 探讨基于二维均匀性直方图和方向均值滤波的去噪算法对超声图像乳腺肿块的检测与分类的价值。方法 对115例乳腺肿块患者（59例良性和56例恶性）共349个肿块进行超声检查，采用新的去噪算法，对超声图像进行去噪处理，采用双盲法评估，由ROC曲线下面积得出此法的特异性和敏感性。结果 乳腺肿块的超声诊断结果及病理诊断结果符合率明显提高，从原片的61例提高到去噪后图片的81例。当假阳性率为0．356时，去噪后图片的敏感性从87．5％提高至98．2％，ROC曲线下面积从0．843增加到0．955。结论 新的超声图像去噪算法明显地改善了图像质量，提高了乳腺肿块的正确诊断率，减低误诊率，有利于计算机辅助自动识别及分类乳腺肿块。

Objective To evaluate the performance of a computerized detection and cIassification with breast ultrasound （US） images using the novel speckle reduction algorithm based on two-dimensional textural homogeneity histogram and directional average filters. Methods Three hundred and forty-nine US images of 115 cases were analyzed including 59 benign lesions and 56 malignant lesions. By using the novel speckle reduction algorithm, the speckles on breast US images were removed. The original and speckle- reduced images were assessed by radiologists using double blind method. The diagnostic sensitivity and specificity were calculated by the areas under the receiver operating characteristic（ROC） curves. Results The breast lesions which can be diagnosed definitely increased from 61 cases of the original images （32 malignant cases and 29 benign cases） to 81 cases of the speckle-reduced images （43 malignant cases and 38 benign cases）. The sensitivity could be raised from 87.5 % to 98.2 % at 0. 356 false-positive detections per image for this detection-plus-classification scheme, and the area under the ROC curve of diagnosis also increased from 0. 843 to 0. 955. Conclusions The novel speckle reduction algorithm was proposed and it can increase the diagnostic accuracy and decrease the rate of missing and misdiagnosis of breast lesions greatly.