Investigating Predictors of Radiation Pneumonitis Complications Caused by Radiation Therapy for Breast Cancer

• Main Authors: Nian-Yuan Song, 宋念遠
• Other Authors: Tsair-Fwu Lee
• Format: Others
• Language: zh-TW
• Published: 2014
• Online Access: http://ndltd.ncl.edu.tw/handle/5322vx

碩士 === 國立高雄應用科技大學 === 電子工程系碩士班 === 102 === Purpose: The aim of this study was to develop a multivariate logistic regression model with select the predictor by least absolute shrinkage and selection operator (LASSO) to make valid predictions about the incidence of radiation pneumonitis among breast cancer patients after radiation therapy. Materials and methods: Eighty-seven patients with breast cancer were analyzed, and evaluation time point is defined in 12 weeks (3months) after radiation therapy. Data compilation used missing value and outlier analysis, after removing the patients with bilateral breast cancer and metastases of cancer. Assessment methods of pneumonitis included (1) CT image was defined significant pneumonia if it was grade ≥1 (RP_CT) (2) clinical symptoms was defined significant pneumonia if it was grade ≥1 (RP_Sym) (3) CT imaging and clinical symptoms occur simultaneously (RP_CT+Sym). Using LASSO with cross-validation to select the predictor. Three learning methods of logistic regression, artificial neural networks and support vector machines were used to build predictive models. Performance of each learning method was assessed by goodness of fit test and predictive performance test. Results: Four predictors were selected by LASSO for CT image grade ≥1: treatment energy, BMI, chemotherapy, ipsilateral V13. Five predictors were selected for clinical symptoms grade ≥1: age, N stage, BMI, treatment energy, supraclavicular fossa. Three predictors were selected for CT imaging and clinical symptoms occur simultaneously: age, BMI, treatment energy. Three learning methods established predictive models were through with goodness of fit test. In predictive performance test, logistic regression models for in assessment methods of AUC values were 0.81, 0.82, 0.79, respectively. Artificial neural networks models in the three assessment methods of AUC values were 0.81, 0.84, 0.88, respectively. Support vector machine prediction model in three assessment methods of AUC values were 0.79, 0.82, 0.84, respectively. Conclusions: BMI and treatment energy are important factors in causing radiation pneumonitis. At three prediction models, artificial neural networks have the best predictive performance.