Research On Lymphovascular Invasion And Lymph Node Metastasis Prediction In Rectal Cancer Based On Radiomics

The incidence and mortality rate of rectal cancer are increasing year by year.Due to the low early diagnosis rate of rectal cancer,most patients have missed the best time for treatment at the time of surgery,resulting in extremely poor prognosis for patients.Lymphovascular invasion and lymph node metastasis are important factors for early diagnosis and assessment of prognosis in rectal cancer.Therefore,preoperative evaluation of lymphovascular invasion and lymph node metastasis status in patients with rectal cancer is crucial.By extracting high-throughput features from medical images and converting medical imaging problems into problems that can be solved using computers,Radiomics has promising applications in clinical aid diagnosis of tumors and development of treatment plans.Based on the Radiomics theory,this paper proposes a classification model for preoperative prediction of lymphovascular invasion and lymph node metastasis status in rectal cancer patients,aiming at the two clinical problems that it is difficult to determine whether lymphovascular invasion and lymph node metastasis in rectal cancer patients by preoperative imaging examination.Fat suppression-T2 weighted imaging(FS-T2WI)images in the paper were obtained from Henan Provincial People’s Hospital.The specific workflow is:The image data are firstly pre-processed with the region of interest outlining,image denoising,wavelet transform,and the clinical data were statistically analyzed.Then three classification models were constructed based on the image data and clinical data:three algorithms of U-test,elastic net,and support vector machine recursive feature elimination are used to filter the image features and obtain the most important features to construct the image model;construction of clinical models using clinical features obtained by screening with a logistic regression algorithm using the Akaike information criterion as an evaluation indicator;combining clinical features with the predictive probabilities of the imaging model and using features filtered by a logistic regression algorithm to construct the combined model.Finally,the prediction performance of the classification model is comprehensively evaluated by AUC,accuracy,specificity,and other evaluation metrics.The experimental results showed that the classification model proposed in this study had good prediction performance.In the classification model for predicting whether the vasculature was invaded,the AUC values of the combined model was 0.954 and 0.909 on the training and test sets,respectively,which had better prediction performance compared to 0.766 and 0.678 for the clinical model.In the classification model for predicting whether the lymph nodes are metastatic,the AUC values of the imaging model were 0.801 and 0.738 on the training and test sets,respectively,compared with 0.617 and 0.557 for the clinical model,which performed better.The calibration curves designed in the experiment showed the validity of the classification model,the decision curves indicated the clinical utility of the classification model.