基于CT的纹理分析预测甲状腺乳头状癌BRAF^(V600E)突变

CT-Based Texture Analysis for Predicting BRAF^(V600E) Mutation in Papillary Thyroid Carcinoma

目的探讨基于CT的纹理分析是否能预测甲状腺乳头状癌BRAF^(V600E)突变情况。资料与方法回顾性分析经手术证实并有BRAF^(V600E)突变结果的甲状腺乳头状癌52例。运用MaZda软件对患者的术前CT增强扫描静脉期图像进行纹理分析,用费希尔参数法(Fisher)、最小分类误差与最小平均相关系数法(POE+ACC)和互信息测度法(MI)3种特征选择算法进行最佳参数筛选,再使用软件内B11程序中的主要成分分析(PCA)、线性判别分析(LDA)和非线性判别分析(NDA)方法对最佳参数进行降维、分类,得到BRAF^(V600E)的误判率R。采用ROC曲线进一步分析不同纹理分析方法对BRAF^(V600E)突变的预测性能。结果BRAF^(V600E)阳性32例,阴性20例。3种特征选择算法中,NDA误判率均最低,均在10%以下,具有很高的诊断性能;其次为LDA,多种组合误判率<20%,诊断性能较好;PCA的误判率最高,均>30%,诊断性能较低。但是3种特征选择算法未表现出明显的优势或劣势。误判率最低的组合为POE+ACC+NDA,ROC曲线下面积(AUC)为0.969;误判率最高的组合为Fisher+PCA,AUC为0.413。结论基于术前CT纹理分析预测甲状腺乳头状癌患者BRAF^(V600E)突变具有一定的可行性,但还需进一步研究证实。

Purpose To investigate whether CT-based texture analysis could predict the BRAF^(V600E)mutation in patients with papillary thyroid carcinoma(PTC).Materials and Methods A total of 52 patients with PTC confirmed as BRAF^(V600E)mutation by surgery were retrospectively analyzed.Texture analysis of the preoperative venous phase contrast-enhanced CT images were performed in patients with PTC by MaZda software,and the Fisher,POE+ACC and MI feature selection algorithms were used to select the best texture feature parameters,then the dimension of best parameters was further reduced and classified via PCA,LDA and NDA in the B11 program,obtaining the error rate R of BRAF^(V600E).ROC curve was used to analyze the prediction performance of BRAF^(V600E)mutation via different texture analysis methods.Results Of 52 patients with PTC,32 patients were BRAF^(V600E)positive and 20 patients were negative,respectively.NDA method showed the lowest error rate R(all less than 10%),with high diagnostic performance;the second was LDA method,and the most combination error rate R were less than 20%,showing good diagnostic performance;the PCA method presented the highest error rate R of>30%,and its diagnostic performance was decreased.However,there were no obvious significant differences in advantages and disadvantages among these three feature selection algorithms.The combination of POE+ACC+NDA showed the lowest error rate R,the AUC was 0.969;the combination of Fisher+PCA was presented the highest error rate,and the AUC was 0.413.Conclusion It is feasible to predict BRAF^(V600E)mutation in patients with PTC based on preoperative CT imaging,but further research is needed.