| Purpose: To set up a predictive model based on the radiomics features of nonsmall cell cancer(NSCLC)to predict brain metastases.Methods: A total of 230 pathologically confirmed NSCLC patients were collected from Zhongda Hospital affiliated to Southeast University from January 2014 to February 2020.Before their initial vist,they hadn’t received any treatment.Collecting the chest enhanced CT images,which were used to extra and select radiomics features and construct the radiomics score.Clinical and CT information were collected,including Age,gender,smoking history,histopathological classification,anatomical site classification,T stage,N stage,overall stage,whether Surgery,chemotherapy,radiation during the follow-up period,Carcinoembryonic antigen(CEA),neuronspecific enolase(NSE),cytokeratin 19 fragment(CYFRA21-1)and maximum tumor diameter of the CT images.Patients were randomly divided into training and validation cohorts according to a 7:3 ratio.ITK-SNAP was used to manually outline region of interest.Python was used to extract and select radiomics features.Logistic regression method was used to create the clinical model,radiomics model and comprehensive model of the clinical and radiomics features in the training cohorts.These models were assessed in the testing cohorts.Delong test was used to test the difference of AUC values between different models.Results: In total,161 patients were included in the training cohort(54 pati ents with brain metastases,107 patients without brain metastases)and 69 patie nts in the test cohort(23 patients with brain metastases,46 patients without bra in metastases).Histopathological classification,Chemotherapy,NSE and CYFRA21-1 were independent correlation factors for brain metastases in the clinical m odel.Four radiomics features,original＿glrlm＿Short Run High Gray Level Emphasis,l og-sigma-2-0-mm-3D＿firstorder＿Maximum,log-sigma-3-0-mm-3D＿firstorder＿Maxi mum and wavelet-HLL＿glszm＿Zone Entropywere included in the assessment of t he radiomics score.The discrimination performance of the integrated model in the training and test cohorts was significantly better than the clinical and radio mics model,with the respective AUC of the model in training cohort and test cohort was 0.886(95%CI:0.832-0.940),0.911(95%CI:0.826-0.996).Conclusion: In this research,it was found that the integrated model which was consisted of histopathological classification,Chemotherapy,NSE,CYFRA21-1 and radiomics score had a better predictive effect on brain metastases than the clinical and radiomics model.The integrated model is most suitable for predicting brain metastases in NSCLC. |
