| Summary: | 碩士 === 國立高雄應用科技大學 === 電子工程系碩士班 === 103 === Purpose: Using Mahalanobis-Taguchi System and multivariate analysis to predict the risk factors of hybrid intensity-modulated radiotherapy-induced dermatitis in breast
cancer patients.
Materials and methods: Eighty-one patients with breast cancer were analyzed. Skin reaction was considered high-grade if it was grade ≥ 2 (defined as Radiation Therapy
Oncology Group, RTOG). The missing data and outlier analysis were used for data compilation, excluding the patients with bilateral breast cancer and metastases of cancer. The method of Mahalanobis-Taguchi system and multivariate analysis were used to select the predictive factors of radiation skin dermatitis. Finally, the performance of each learning method was assessed by goodness of fit test and predictive performance test.
Results: Using Mahalanobis-Taguchi system and multivariate analysis to select predictors of breast cancer patients for the induced dermatitis after radiation therapy were: Planning target volume (PTV_100)、V30、T stage、Age and PTV_100、V30、T-Boost. For the area under the ROC curve (AUC) values of goodness of fit test, the
logistic regression models established by the Mahalanobis-Taguchi system was 0.84 and the multivariate analysis was 0.81.
Conclusions: PTV_100 and breast skin dose volume histograms (V30) are the two important factors causing radiation skin dermatitis. This study compared the Mahalanobis-Taguchi system and multivariate analysis, we found that the Mahalanobis-Taguchi system had the better predictive performance than the multivariate analysis .
|
|---|
