Prediction of multi-parameter MRI radiomics based on machine learning in Ki-67 expression in laryngeal squamous cell carci-noma
Objective To explore the value of machine learning-based multi-parameter MRI radiomics in predicting Ki-67 expression in laryngeal squamous cell carcinoma(LSCC)before surgery.Methods The clinicopathological and MRI data of 271 patients with LSCC confirmed by pathological examination in the Fifth Hospital Affiliated to Wenzhou Medical University(Center 1)from January 2016 to February 2024 and Lishui People's Hospital(Center 2)from Januany 2019 to December 2023 were retrospectively analyzed.A total of 189 patients from Center 1 were randomly divided into a training set of 132 cases and a validation set of 57 cases in a ratio of 7∶3;another 82 patients from Center 2 were used as an external test set.The radiomic features of lesions were extracted from T2WI and contrast-enhanced T,weighted imaging(CE-T1WI)images.The optimal features were obtained through dimensionality reduction,and six machine learning classifiers were established.The classifier with the highest average AUC value in the validation set and the external test set was selected as the best radiomics model,and its results were converted to a radiomics score(Rad-score).Clinical features with P＜0.05 in the univariate analysis were included in the multivariate logistic regression analysis to obtain the risk factors associated with high Ki-67 expression for establishing a clinical model.Finally,a combined model was constructed based on clinical features and radiomic features,and a nomogram was drawn.The ROC curve were used to evaluate the performance of different models in predicting Ki-67 expression.Results Twelve optimal radiomics features were obtained from T2WI and CE-T1WI images.In the validation set and external test set,the AUC ranges of the six machine learning classifiers were 0.647-0.829,and 0.664-0.803,respectively.Among them,random forest had the best predictive performance(average AUC of 0.816).The clinical T stage,MRI-reported lymph node status,and Rad-score were further combined to establish a nomogram.The ROC curve results showed that the AUC of the nomogram in the training set,validation set,and external test set were 0.923,0.870,and 0.822,respectively.Conclusion Multi-parameter MRI radiomics based on machine learning has good predictive value for the expression level of Ki-67 in LSCC patients,and the nomogram established by further combining clinical features can better improve performance.
国家科技期刊平台
NSTL
万方数据
