| 彭媛媛,任翠萍,程敬亮,文宝红.瘤体最大层面ADC图纹理分析鉴别鼻腔鼻窦小圆与非小圆细胞恶性肿瘤[J].中国介入影像与治疗学,2021,18(1):37-41 |
| 瘤体最大层面ADC图纹理分析鉴别鼻腔鼻窦小圆与非小圆细胞恶性肿瘤 |
| ADC image texture analysis of maximum tumor level in differential diagnosis of small round and non-small round cell malignant tumor of nasal and paranasal sinus |
| 投稿时间：2020-06-06 修订日期：2020-11-30 |
| DOI：10.13929/j.issn.1672-8475.2021.01.009 |
| 中文关键词: 鼻咽肿瘤 纹理分析 磁共振成像 |
| 英文关键词:nasopharyngeal tumors texture analysis magnetic resonance imaging |
| 基金项目:河南省医学科技攻关计划（联合共建）项目（LHGJ20190157）。 |
|
| 摘要点击次数: 1122 |
| 全文下载次数: 471 |
| 中文摘要: |
| 目的 观察基于瘤体最大层面ADC图纹理分析鉴别鼻腔鼻窦小圆细胞恶性肿瘤（SRCMT）与非小圆细胞恶性肿瘤（non-SRCMT）的价值。方法 50例SRCMT及55例non-SRCMT患者均接受DWI并经病理确诊。采用MaZda软件于ADC图上手动勾画瘤体各层面，并以最大层面作为ROI，分别以Fisher系数（Fisher）、聚类偏差概率结合平均关联系数（POE+ACC）、协同信息（MI）及联合应用三者（MI+PA+F）筛选鉴别诊断价值最佳的特征参数集，并分别对其进行人工神经网络（ANN）分类及1-邻近分类测试，计算不同特征筛选及分类方法鉴别SRCMT与non-SRCMT的错判率。对错判率最小的特征筛选及分类方法所得参数进行对比分析及受试者工作特征（ROC）曲线分析，筛选SRCMT与non-SRCMT组间差异有统计学意义的参数，并评估其诊断效能。结果 对Fisher、POE+ACC、MI、MI+PA+F筛选特征集进行A-NN分类所得错判率均小于1-邻近分类错判率，采用MI+PA+F筛选特征并进行ANN分类错判率最小，为7.62%（8/105）。采用MI+PA+F方法得出的30个最佳纹理参数中，16个在SRCMT与non-SRCMT组间差异有统计学意义（P均<0.05），对应AUC为0.645~0.722。结论 基于瘤体最大层面ADC图纹理分析可用于鉴别鼻腔鼻窦SRCMT与non-SRCMT；MI+PA+F与A-NN组合错判率最小，可获得具有较高诊断效能的纹理参数。 |
| 英文摘要: |
| Objective To investigate the value of ADC image texture analysis of maximum tumor level for differentiating small round cell malignant tumor (SRCMT) and non-SRCMT. Methods Data of 50 SRCMT and 55 non-SRCMT patients were retrospectively analyzed. All patients underwent DWI examination, and the diagnoses were then confirmed by pathology. MaZda software was used to manually delineate tumor layer on ADC image and elect maximum level as ROI. The optimal feature parameters set of texture analysis were obtained with the methods of Fisher coefficient (Fisher), probability of classification error and average correction coefficient (POE+ACC), mutual information (MI) and combination of the above three methods (MI+PA+F), respectively. Then 1-nearest neighbor classification and the artificial neural network (ANN) classification tests were respectively performed on these parameters to obtain the misclassification rate of SRCMT and non-SRCMT. Contrastive analysis and ROC curve analysis were performed on the parameters obtained by feature screening and classification methods with the lowest misclassification rate, in order to find out the parameters with statistically different between SRCMT and non-SRCMT groups and evaluate their diagnostic efficacy. Results Among the feature sets screened by Fisher, POE+ACC, MI and MI+PA+F, the misclassification rate obtained by ANN classification was lower than that by 1-nearest neighbor classification, and the misclassification rate obtained by MI+PA+F screening and A-NN classification was the lowest, which was 7.62% (8/105). Among 30 optimal texture parameters obtained by MI+PA+F method, 16 were statistically different between SRCMT and non-SRCMT (all P<0.05), and their corresponding AUC were 0.645 to 0.722. Conclusion The texture analysis of ADC images at the maximum tumor level might be used to differentiate SRCMT and non-SRCMT of nasal and paranasal sinus. Combination of MI+PA+F and A-NN had the minimum misclassification rate, and the obtained texture parameters had relative high diagnostic efficiency. |
| 查看全文 查看/发表评论 下载PDF阅读器 |
| 关闭 |
