Analysis Of Therapeutic Effect For High Dose Imrt Treated Primary Or Metastatic Lung Cancer And The Svm Classifier For Prediction Of Lung Radiated Injury

objective: First part:the therapeutic effect and radiation induce lung injury by high dose IMRT in primary or metastatic lung cancer were analyzed.Second part:a series of patients with clinical and dosimetry data were selected,and modeling SVM classifier to predict radiation induced lung injury.Method and Material:413 patients with primary lung cancer and metastatic lung tumor and recurrent lung cancer after surgery were selected from Sichuan Province Cancer Hospital and Institute. Available 112 patients with acute radiation induced pneumonitis(ARP) and radiation induced pulmonary fibrosis(RPF) evaluated by CTCAE version.3, and ≥6 months’ follow up were needed. In the first part, survival analysis was used for high dose IMRT treated patients, and the therapeutic effect and radiation induce lung injury were compared for “ART and non-ART” or “ART alone and ART with dose painting”. In the second part, modeling a SVM model to predict the radiation induced lung injury based on the basic clinical data and dose-volume parameters, and the performance of LR or rcs-LR model were compared with SVM model finally.Result:First part: Kaplan-Miere analysis for 73 cases with primary NSCLC show that 6 months and 1,2 years local control and overall survival rates were 97.2%, 84.3%, 69.2% and 98.6%, 82.5%, 44.8%, respectively.The median survival time was 20.1 months(95%CI,14.17~26.03). Logistic regression show that age above 60 years and central located cancer were risk factor for ≥Grade 2 ARP in patients with primary NSCLC. And the Cox proportional risk regression show that patients do not use the target therapy or with DM were risk factor for local control rate in primary NSCLC cohort. Significant statistics in survival or adverse effect were not found in “ART and non-ART” or “ART alone and ART with dose painting”, respectively.But comparing with non-ART or ART without Dose painting,high dose BED with ART or ART+Dose painting method might improve the curative effect potentially with limited radiation induced lung injury.Kaplan-Miere analysis for 39 cases of metastatic lung cancer with 51 lesions show that 6months and 1,2 years local control and overall survival rates were 100%,85.1%,79.8% and 100%,85.2%,50.0%,respectively.The median survival time was 22.5 months. And the 6months and 1,2 years local control rate for 51 lesions were 98.1%,82.3%,69.4%,the median local progression free survival time was 40.9 months(95%CI,NA).Second part: The median age was 61 years old(20 ~ 92 years), the median follow-up time was 16.7 months(6-97.7 months) in 158 cases with primary orrecurrent or metastatic lung cancer. Univariate analysis show that age, tumor position(peripheral or central located), GTV, PTV, GTV/ipsilateral normal lung volume, GTV/whole lung volume, PTV/ipsilateral normal lung volume, PTV/whole lung volume, ipsilateral lung Vmean, V5, V10, V20, V30, V40, MLD, D5, D10, D20, D25, D30, D98,contralateral lung V5, D98,total tumor diameters before therapy D0 were impact ≥G2 ARP significantly; Tumor clinical stage, pathology type, dose range per fraction, tumor position(peripheral or central located), GTV, PTV, GTV/ipsilateral normal lung volume, GTV/whole lung volume, PTV/ipsilateral normal lung volume, PTV/whole lung volume, ipsilateral lung V30,V40,D5,D10,BED(α/β=3), EQD2(α/β=3), total tumor diameters before therapy D0 were significant impact ≥ G2 RPF statistically. Multivariate analysis show that i D10,tumor position(peripheral or central located) were independent prognostic factors of ≥G2 ARP; PTV, tumor position(peripheral or central located) and fraction type(Qd and Bid) were independent prognostic factors for ≥G2 RPF. Support vector machine SVM predict effectively compared with the traditional logistic regression model(adjusted C-index were 0.67 for ARP and 0.74 for RPF, respectively)or restricted cubic spline fixed LR model(adjusted C-index were 0.68 for ARP and 0.70 for RPF, respectively),and SVM classifier for acute radiation induced pneumonitis show that 83% in sensitivity,74% in specificity,78% in classification accuracy, and the AUC is 0.79; Radiationinduced pulmonary fibrosis SVM classifier showed that 75% in accuracy, 69% in sensitivity,80% in specificity, and AUC is 0.74.Both SVM model provide much more reliable results.Conclusion:1.High dose IMRT(BED≥60Gy) could treat primary NSCLC effectively with tolerated radiation induced lung injury. 2.Compare with conventional RT planning, ART or ART with dose painting in primary NSCLC patients might potentially improved survival benefit with tolerated adverse effect, long term research is valuable. 3.Age above 60 yrs and central located NSCLC were independent risk factor for Grade 2+ ARP. 4.Characteristics chosen from variance selection algorithm in SVM could predict radiation induced lung injury effectively. 5. Compared with the traditional logistic regression model and restricted cubic spline fixed LR model, the power of prediction and classification in support vector machine(SVM) showed quite effective performance. The better classification and prediction of pulmonary injury could SVM archived, and more, could SVM provide much more reliable results in clinical decision also.