Preoperative Diagnostic Value Of Multi-parametric MRI Radiomics For Lymph Node Metastasis Of Rectal Cancer

Objective:To explore the value of multi-parametric MRI radiomics model based on machine learning for preoperative prediction of rectal cancer lymph node metastasis.Methods:A total of 140 rectal cancer patients who underwent radical surgical resection with pathological results between January 2016 and March 2017 were included in this retrospective study.The patients were randomly divided into a training set(98 cases)and a validation set(42 cases).All the patients underwent rectal MRI scan Within 2 weeks before the surgery.Using Intelli Space Discovery to map the volume of interest(VOIs)of primary tumors and mesorectal fasciaon on T2 WI sequences,using Philips radionics tools to extract radiomics features of VOIs.For each ROI,a total of 1227three-dimensional(3D)based radiomic features were extracted.For each patient,we integrated all of the 2454 Radiomics features from two VOIs together.Spearman correlation analysis was used to evaluate the correlation between theextracted Radiomics features and lymph node metastasis of rectal cancer,and then the least absolute shrinkage and selection operator(LASSO)was used for further feature dimensionality reduction.In the modeling stage,we investigated 23 classification methods based on machine learning for training and prediction of the radiomics model.Multivariate models were trained on the training cohort through five-fold cross-validation and their performance was evaluated on the validation cohort using the area under ROC curve(AUC),accuracy,specificity,sensitivity,and F1 score.Results:There were no significant differences in clinical and pathological features between the training and validation groups(P>0.05).A total of 1227 three-dimensional(3D)based radiomics features were extracted for each VOI and a total of2454 radiomics features were extracted from each patient's two VOIs.A total of 21 radiomics features were retained by Spearman correlation analysis and the LASSO algorithm to construct a predictive model.In the 23 machine learning models,the Ridge Classifier found the best predictive performance.The average AUC of the training set,the fivefold cross-validation set,and the test set are 0.905,0.803,and 0.785,respectively.The accuracy was 82.7%,79.5% and73.8%,respectively.At the same time,the mean specificity,sensitivity,and F1 scores were 0.79,0.667 and 0.785,respectively.Conclusion:The multi-parameter MRI radiomics model based on machine learning is a powerful tool for individualized and noninvasive prediction of lymph node metastasis with high diagnostic performance.