| Purpose: Meningioma is the most common primary intracranial tumor.Accurately predict the surrounding tissues invasion of meningioma before surgery,including brain invasion,bone invasion and sinus invasion,which would contribute to ameliorate clinical decision-making,predicting meningioma grading and prognosis,and are the key steps to individualized treatment of meningioma.This study aimed at the clinical needs of preoperative prediction of brain invasion,bone invasion and sinus invasion of meningioma,and constructed a radiomics prediction model based on magnetic resonance imaging(MRI).Materials and methods: 1.A total of 1728 patients with pathologically confirmed meningioma were collected from two clinical medical centers retrospectively(training cohort: 1070 cases in Beijing Tiantan Hospital;external validation cohort: 658 cases in Lanzhou University Second Hospital).The radiomic features were extracted from T1-weighted post-contrast(T1C)and T2-weighted(T2)images respectively.The least absolute shrinkage and selection operator was used to select the most informative features of different modalities,and the support vector machine algorithm was used to construct different models(T1C,T2,T1C+T2,clinical risk factors + T1C+T2)to predict the risk of brain invasion.Then,the optimal model and clinical risk factors were enrolled to build a nomograph.The receiver operating characteristic curve(ROC)AUC,accuracy,specificity and sensitivity were calculated,and the calibration curve and clinical decision curve analysis were used to validate stability and the clinical value of nomograph.2.A total of 490 patients with meningiomas confirmed by pathology in Lanzhou University Second Hospital(448 cases of WHO grade ?,38 cases of grade ?,4 cases of grade ?)were enrolled.All patients were randomly divided into training(n= 343)and test(n = 147)datasets at a 7:3 ratio.For each patient,1227 radiomic features were extracted from T1 C and T2,respectively.Spearman's correlation and LASSO regression analyses were performed to select the most informative features.Subsequently,a 5-fold cross-validation was used to compare the performance of different classification algorithms,and logistic regression was chosen to predict the risk of bone invasion.3.A total of 348 patients with parasinus meningiomas confirmed by pathology were enrolled Lanzhou University Second Hospital(291 cases of WHO grade ?,50 cases of grade ?,7 cases of grade ?)retrospectively.All patients were randomly divided into training(n = 244)and test(n = 104)datasets at a 7:3 ratio.For each patient,1037 radiomic features were extracted from T1 C and T2,respectively.The Pearson correlation analysis,LASSO and logistic regression were used to select the most informative features of different modalities,and the logistic regression was used to construct different models(T1C,T2,T1C+T2)to predict the risk of venous sinus invasion and the optimal model was selected.The AUC,accuracy,specificity and sensitivity of ROC were calculated,and the calibration curve and clinical decision curve analysis were used to validate stability and the clinical value of the model.Results: 1.Sixteen features were screened from 3190 radiomic features,which were significantly related to brain invasion(8 from T1 C,8 from T2).Among the four models,the radiomics model constructed by T1 C and T2 showed good discrimination ability.The areas under the curves(AUCs)in the training cohort and the external validation cohort was 0.855(0.829-0.882)and 0.796(0.747-0.845),and the sensitivity was 80.3% and 79.0%,respectively.While the clinicoradiomic model derived from the fusing MRI sequences and sex resulted in the best discrimination ability for risk prediction of brain invasion,the AUCs with of 0.857(95% CI,0.831-0.887)and 0.819(95% CI,0.775-0.863)and sensitivities of 72.8% and 90.1% in the training and validation cohorts,respectively.Compared with radiomic model,the performance of clinicoradiomic model improved by 2.42% in discrimination ability,P=0.0142,which was a significant difference.The calibration curve analysis showed that the probability of brain invasion predicted by the fusion model was good agreement with the actual brain invasion.The clinical decision curve analysis also showed that the fusion prediction model was the best net benefit.2.Eight features extracted from T1 C and T2 images,which were closely related to bone invasion.The AUC values of the T1 C,T2,and fusion T1 C and T2 models established by logistic regression were 0.714(0.660-0.768),0.679(0.621-0.737),0.722(0.668-0.776)and 0.715(0.632-0.798),0.686(0.599-0.7742),0.713(0.628-0.798)in the training cohort and validation cohort,respectively.The radiomic models derived from T1 C alone or a combination of T1 C and T2 had good performance in predicting risk of bone invasion.3.Seven features were screened from 2074 radiomic features,which were significantly related to venous sinus invasion(3 features from T1 C,4 features from T2).Among these models,the radiomics model constructed by T1 C and T2 showed good discrimination ability.The AUCs in the training cohort and the validation cohort was 0.748(0.681-0.815)and0.740(0.645-0.835),and the sensitivity was 90.8% and 94.6%,respectively.The calibration curve analysis showed that the probability of venous sinus invasion predicted by the fusion model was good agreement with the actual venous sinus invasion.The clinical decision curve analysis demonstrated that the fusion prediction model was the best net benefit.Conclusion: 1.The clinicoradiomic model incorporating sex information and the radiomic signatures showed great performance and high sensitivity in predicting brain invasion with meningioma,and validated in the external validation cohort,which can be used to guide clinical surgical decision-making,predict pathological grade and prognosis.2.The model constructed and validated by the radiomics features of MRI may effectively predict the bone invasion of meningiomas,and can be used as a new potential tool for making the surgical plans and predicting the prognosis.3.The radiomic model constructed in this study had a good predictive performance for venous sinus invasion in meningioma,and the model was verified by the accuracy,sensitivity,specificity,AUC value,calibration curve and decision curve analysis in the internal validation set.The results showed that the model had good predictive ability and generalization ability.4.The radiomic model can be used as a potentially valuable and easy-to-operate tool to assess comprehensively and accurately the invasion of important tissue structures such as brain tissue,bone tissue,and venous sinuses around meningioma before surgery.These studies contributed to achieving surgical risk stratification of patients in meningioma,guiding intraoperative decision-making,reducing postoperative recurrence rate,and attaining ultimately individualized and precise diagnosis and treatment of patients in meningioma,in order to ensure the maximum survival benefit of patients. |
PDF Full Text Request