Objective To develop and validate a radiomics nomogram model combined with Machine Learning to pre-dict short-term outcomes in spontaneous intracerebral hemorrhage.Methods We retrospectively included 289 patients with acute ICH in our hospital between October 2019 and October 2021,and divided them into a training cohort and an in-ternal validation cohort at a ratio of 7∶3.Based on the Least Absolute Shrinkage and Selection Operator(LASSO)algo-rithm,the optimal parameters were debutted by 5x cross-validation method,and the prediction models were built by combi-ning 5 machine algorithms.The area under curve(AUC)of the receiver operating characteristic(ROC)curve were used to evaluate these models.Multivariate Logistic regression analysis of the training set was used to construct a comprehensive prediction model of imaging and clinic,and the nomogram were drawn.we used data from an external hospital of 163 pa-tients served as an independent external test cohort to validate the model.Results By comparing the performance of five machine learning algorithms,the random forest(RF)model showed the best performance(AUC=0.83).The AUC of ra-diomics nomogram model combined with Machine Learning in training cohort,internal validation cohort,external validation cohort were 0.88,0.86,0.86,respectively.Calibration curves showed satisfactory effect in both training and external co-horts(both P＜0.05),whereas in internal cohort there were less consistency(P＜0.05).Conclusion Radiomics nomo-gram model combined with Machine Learning is an effective tool to provide personalized risk assessment of short-term out-comes for ICH patients,in which the RF algorithm model performs best.
NETL
NSTL
万方数据
