计算机辅助鉴别乳腺良、恶性肿块的临床价值

Clinical value of computer-aided algorithm on radiologist＇s diagnosis of masses using breast ultrasound imaging

目的提出～种基于灰度级二维直方图的计算机辅助分割算法，对乳腺高频超声图像中的肿块进行自动识别和分割处理，旨在提高乳腺良、恶性肿块的检出准确率。方法采集100个乳腺肿块二维超声图像共466张（原片），应用计算机软件对原片进行分割处理，得到分割后图片。超声医师采用双盲法分别根据原片和分割后图片中的超声征象进行良、恶性判断，运用受试者工作曲线（ROC曲线）计算曲线下面积（A），比较前后两次诊断结果，分析图片处理前后诊断结果的差异性。结果处理后的图片中乳腺肿块的边缘、钙化等信息明显突出。超声医师对良、恶性肿块的确诊率明显提高。当特异性为74．31％时，诊断敏感性由基于原片的70．32％提高到图片分割后的90．52％。ROC曲线下面积由分割前的80．8％上升到90．5％，差异有统计学意义（P〈0．01）。结论此分割算法能明显优化乳腺肿块的边缘信息，较好地突显肿块内微钙化，在一定程度上降低漏诊率和误诊率，提高乳腺良恶性肿块的确诊率。

Objective To propose a novel computer-aided segment algorithm, and retrospectively investigate its effect on radiologist＇s sensitivity and specificity for discriminating malignant masses from benign masses basing on ultrasound images. Methods Four hundred and sixty-six images of 100 masses obtained by conventional ultrasound were processed by a novel segment algorithm. Radiologists who were blind to the histology results were invited to analyse the original and computerized images respectively. The sensitivity and specificity were calculated by means of a binary outcome using receiver operating characteristic（ROC） analysis. Results By using the segment algorithm, the quantity of the image was obviously improved,especially the margin and calcification of the masses. The diagnostic performance of the radiologists was also improved. The sensitivity rose from 70.32% to 90. 52% while the specificity was 74.31%. The area under the ROC curve increased significantly （P 〈 0.01） from 80.8% to 90.5%. Conclusions The proposed segmentation algorithm can improve the diagnosis performance of the radiologists by meliorating the quality of the ultrasound image.