中文摘要：

目的：本文尝试设计一款基于高清晰度 CT （HDCT ）图像的孤立性肺结节（SPN ）计算机辅助诊断系统（CADS），以提高恶性 SPN 的检出率，使诊断更加客观、科学。方法收集经临床病理证实的孤立性肺结节120例，包括恶性肿瘤、良性肿瘤、结核和炎性假瘤，随机抽取60例作为实验集，60例作为验证集。实验集 HDCT 图像经图像预处理、感兴趣区域（ROI）基于标记的分水岭算法分割和 ROI 纹理特征参数提取，对获得的5项纹理特征参数做统计学处理，将统计结果应用于系统以对 SPN 做良恶性分析并给出提示信息。验证集 HDCT 图像输入系统后，对比系统预测结果与主任医师和住院医师预测结果来评价系统的可靠性。结果对比度、相关性、熵、平稳度和二阶矩 t 检验 P 值分别为0.000、0．002、0．914、0．295、0．002。对比度、相关性和二阶矩良性区间分别为（903，2003）、（2．76，3．48）、（0．01，1．54），恶性区间分别为（502，898）、（3．49，3．71）、（1．79，29．86）。系统、主任医师和住院医师的敏感度分别为83．3％、93．3％、76.7％，假阳性率分别为13．3％、16．7％、26．7％，正确率分别为85％、88．3％、75％。结论基于标记的分水岭算法对与胸壁或纵隔粘连的结节及磨玻璃病变等均可以较好地将其分割提取出来。对比度、相关性和二阶矩有统计学意义。系统在预测 SPN 良恶性具有较高的敏感性和准确性及最低的假阳性。 CAD 在 SPN 良恶性诊断具有一定的临床使用价值，本系统可以辅助临床医师诊断 SPN 良恶性。

英文摘要：

Objective In this paper ,we try to design a computered‐aided diagnostic system （CADS） of solitary pulmona‐ry nodule （SPN） based on high definition CT （HDCT ） images in order to improve the detection rate of malignant SPN and to make the diagnosis more objective and scientific .Methods 120 cases of SPN confirmed by pathology were collected ,in‐cluding malignant tumors ,benign tumors ,tuberculosis and inflammatory pseudotumors .60 cases were randomly selected as the experimental set ,60 cases as the validation set .We acquired five texture feature parameters from HDCT images of the experimental set by images preprocessed ,regions of interest （ROI） segmented by watershed algorithm based on the mark and texture feature parameters of ROI extracted .A statistical method was used to process the five texture feature pa‐rameters .The statistical results were applied in the CAD to differentiate malignant SPN from benign SPN and give messa‐ges .The HDCT images of the validation set were inputed in the CAD and we could evaluate the reliability of the CAD by comparing the forecasts of the CAD ,chief physicians and residencies .Results The results of t test of the five texture fea‐ture parameters were 0 .000 ,0 .002 ,0 .914 ,0 .295 ,0 .002 respectively .The benign ranges of the parameters were respec‐tively （903 ,2003） ,（2 .76 ,3 .48） ,（0 .01 ,1 .54） and the malignant ranges were （502 ,898） ,（3 .49 ,3 .71） ,（1.79 , 29.86） respectively .The sensitivity of the CAD was 83 .3% ,the false positive rate was 13 .3% and the correct rate was 85% .Conclusion The watershed algorithm based on the mark could better segment the nodules adjoined with the chest wall or mediastinum and the ground glass lesions .There was statistical significance in contrast ,correlation and energy .The CAD had higher sensitivity and accuracy and the lowest false positive .The CAD possessed certain clinical value in dif‐ferentiating malignant SPN from benign SPN and could assist clinicians .