逐步判别分析在星形细胞瘤分级中的应用

背景与目的：星形细胞瘤是最常见的神经上皮性肿瘤，其术前分级对治疗和预后影响很大。本研究根据星形细胞瘤相关因素建立逐步判别分析模型，探讨逐步判别分析在星形细胞瘤分级中的应用价值。方法：2008年1月-2009年4月，收治111例首发星形细胞瘤患者。根据磁共振成像（magnetic resonance imaging，MRI）资料按照评分标准分别对每位患者的肿瘤所在部位、平扫T1信号强度、平扫T2信号强度、增强扫描后是否增强、肿瘤周围是否有水肿、肿瘤边界是否清晰、肿瘤囊性情况和肿瘤有无占位效应等进行评分。连同年龄评分以及病理分级评分后，使用Fisher逐步判别分析和Logistic逐步判别分析进行预测，并分析对比两种模型的结果。结果：Fisher逐步判别分析对星形细胞瘤分级的预测准确率为87．7％，灵敏度80．0％，特异度为91．5％，ROC曲线下面积0．942；Logistic判别的预测准确率为84．9％，灵敏度80．0％，特异度为86．8％，ROC曲线下面积0．940；两种模型的准确率差异无统计学意义（P=0．250），ROC曲线下面积差异无统计学意义，Z=0．433（P=0．665）。结论：两种逐步判别分析对星形细胞瘤病理分级预测均有意义，且Fisher判别分析的应用更为简单。

Application of stepwise discriminant analysis for grading of astrocytomas

Background and purpose: Astrocytoma is the most common neuroepithelial neoplasm, and its grading has profound effect on its treatment and prognosis. To investigate the application of stepwise discriminant analysis in grading astrocytomas, this study developed two models of stepwise discriminant analysis according to relevant factors of astrocytoma. Methods: From January 2008 to April 2009, 111 primary astrocytoma patients were enrolled. Each patient was scored based on location, signal intensity on T1WI, signal intensity on T2WI, enhancement, edema, border, cyst or solidness, and mass effect of their magnetic resonance images. With their age score of grading, Fisher stepwise discriminant analysis and the Logistic discdminant were used. The results from the two models were then evaluated and compared. Results: According to Fisher stepwise diseriminant analysis, the predictive accuracy was 87.7% with 80.0% sensitivity, 91.5% specificity and 0.942 area of ROC curve. However, the predictive accuracy of Logistic discriminant analysis was 84.9% with 80.0% sensitivity, 86.8% specificity and 0.940 area of ROC curve. There were no statistically significant differences in terms of accuracy (P=0.250) and areas under ROC curve (Z=0.433, P=0.665) between the two models. Conclusion: Two stepwise discriminant analysis models are meaningful to predict the grading of astrocytoms, and the application of Fisher stepwise discriminant analysis is simpler than the Logistic discriminant analysis.