Receive a weekly summary and discussion of the top papers of the week by leading researchers in the field.

In European journal of radiology ; h5-index 47.0

PURPOSE : To investigate the prediction performance of radiomic models based on multiparametric MRI in predicting the meningioma grade.

METHOD : In all, 229 low-grade [Grade I] and 87 high-grade [Grade II/III] patients with pathologically diagnosed meningiomas were enrolled. Radiomic features from conventional MRI (cMRI), ADC maps and SWI were extracted based on the volume of entire tumor. Classification performance of different radiomic models (cMRI, ADC, SWI, cMRI + ADC, cMRI + SWI, ADC + SWI, and cMRI + ADC + SWI models) was evaluated by a nested LOOCV approach, combining the LASSO feature selection and RF classifier that was trained (1) without subsampling, and (2) with the synthetic minority over-sampling technique (SMOTE). The prediction performance of radiomic models was assessed using ROC curve and AUC of them was compared using Delong's test.

RESULTS : The cMRI + ADC + SWI model demonstrated the best performance without or with subsampling, which AUCs were 0.84 and 0.81, respectively. Following the cMRI + ADC + SWI model, the AUC range of the other models was 0.75-0.80 without subsampling, and was 0.71-0.79 with subsampling. Although the cMRI + ADC model and cMRI + SWI model showed higher AUCs than the cMRI model without subsampling (0.77 vs 0.80, P = 0.037 and 0.77 vs 0.80, P = 0.009, respectively), there was no significant difference among these models with subsampling (0.78 vs 0.77, P = 0.552 and 0.78 vs 0.79, P = 0.246, respectively).

CONCLUSIONS : Multiparametric radiomic model based on cMRI, ADC map and SWI yielded the best prediction performance in predicting the meningioma grade, which might offer potential guidance in clinical decision-making.

Hu Jianping, Zhao Yijing, Li Mengcheng, Liu Jianyi, Wang Feng, Weng Qiang, Wang Xingfu, Cao Dairong


Apparent diffusion coefficient, Machine learning, Magnetic resonance imaging, Meningiomas, Radiomics, Susceptibility weighted imaging