User login
, according to a study published in the European Journal of Radiology.
The model – which combines conventional magnetic resonance imaging (cMRI), apparent diffusion coefficient (ADC) maps, and susceptibility weighted imaging (SWI) – was the best performer of all models tested.
Recent studies have shown that radiomic features from cMRI or ADC maps could build “a robust model to predict the grade of meningioma by using machine learning algorithms,” wrote study author Jianping Hu, MD, PhD, of Fujian Medical University in Fujian, China, and colleagues.
With that in mind, the researchers evaluated the role of radiomic models based on cMRI, ADC maps, and/or SWI in predicting meningioma grade.
Patients and models
The team retrospectively analyzed 514 patients with meningioma who underwent preoperative MRI assessment over a 10-year period. There were 316 patients included in the final analysis, 229 with low-grade (grade I) and 87 with high-grade (grade II-III) meningioma.
Radiomic features from cMRI, ADC maps, and SWI were extracted based on total tumor volume.
Using a nested leave-one-out cross-validation method, the researchers evaluated the prediction performance of various radiomic models, including cMRI, ADC, SWI, ADC plus SWI, cMRI plus ADC, cMRI plus SWI, and cMRI plus ADC plus SWI.
To establish the final prediction model, the researchers used least absolute shrinkage and selection operator feature selection and implemented a random forest classifier that was trained with and without subsampling. The area under the receiver operating characteristic curve (AUC) was used to evaluate the prediction performance of each model.
Results
The model combining cMRI, ADC, and SWI had the best performance in predicting meningioma grade. The AUC of this model was 0.81 with subsampling and 0.84 without subsampling. The other models had an AUC range of 0.71-0.79 with subsampling and 0.75-0.80 without subsampling.
“Our results indicated that [the] multiparametric radiomic model based on cMRI, ADC map, and SWI [tended] to be the best model for the prediction of meningioma grade,” Dr. Hu and colleagues wrote.
Other recent studies have demonstrated that radiomic features from various imaging parameters, such as diffusion weighted imaging and cMRI, can establish robust prediction models for the prediction of meningioma grade, in which AUCs have ranged from 0.63 to 0.91.
While these findings are encouraging, the researchers acknowledged that these data should be interpreted with discretion as the cystic or necrotic areas of tumor were included in the analysis. In addition, the retrospective nature of the study could have introduced selection bias.
No funding sources were reported. The authors reported having no conflicts of interest.
SOURCE: Hu J et al. Eur J Radiol. 2020 Aug 28. doi: 10.1016/j.ejrad.2020.109251.
, according to a study published in the European Journal of Radiology.
The model – which combines conventional magnetic resonance imaging (cMRI), apparent diffusion coefficient (ADC) maps, and susceptibility weighted imaging (SWI) – was the best performer of all models tested.
Recent studies have shown that radiomic features from cMRI or ADC maps could build “a robust model to predict the grade of meningioma by using machine learning algorithms,” wrote study author Jianping Hu, MD, PhD, of Fujian Medical University in Fujian, China, and colleagues.
With that in mind, the researchers evaluated the role of radiomic models based on cMRI, ADC maps, and/or SWI in predicting meningioma grade.
Patients and models
The team retrospectively analyzed 514 patients with meningioma who underwent preoperative MRI assessment over a 10-year period. There were 316 patients included in the final analysis, 229 with low-grade (grade I) and 87 with high-grade (grade II-III) meningioma.
Radiomic features from cMRI, ADC maps, and SWI were extracted based on total tumor volume.
Using a nested leave-one-out cross-validation method, the researchers evaluated the prediction performance of various radiomic models, including cMRI, ADC, SWI, ADC plus SWI, cMRI plus ADC, cMRI plus SWI, and cMRI plus ADC plus SWI.
To establish the final prediction model, the researchers used least absolute shrinkage and selection operator feature selection and implemented a random forest classifier that was trained with and without subsampling. The area under the receiver operating characteristic curve (AUC) was used to evaluate the prediction performance of each model.
Results
The model combining cMRI, ADC, and SWI had the best performance in predicting meningioma grade. The AUC of this model was 0.81 with subsampling and 0.84 without subsampling. The other models had an AUC range of 0.71-0.79 with subsampling and 0.75-0.80 without subsampling.
“Our results indicated that [the] multiparametric radiomic model based on cMRI, ADC map, and SWI [tended] to be the best model for the prediction of meningioma grade,” Dr. Hu and colleagues wrote.
Other recent studies have demonstrated that radiomic features from various imaging parameters, such as diffusion weighted imaging and cMRI, can establish robust prediction models for the prediction of meningioma grade, in which AUCs have ranged from 0.63 to 0.91.
While these findings are encouraging, the researchers acknowledged that these data should be interpreted with discretion as the cystic or necrotic areas of tumor were included in the analysis. In addition, the retrospective nature of the study could have introduced selection bias.
No funding sources were reported. The authors reported having no conflicts of interest.
SOURCE: Hu J et al. Eur J Radiol. 2020 Aug 28. doi: 10.1016/j.ejrad.2020.109251.
, according to a study published in the European Journal of Radiology.
The model – which combines conventional magnetic resonance imaging (cMRI), apparent diffusion coefficient (ADC) maps, and susceptibility weighted imaging (SWI) – was the best performer of all models tested.
Recent studies have shown that radiomic features from cMRI or ADC maps could build “a robust model to predict the grade of meningioma by using machine learning algorithms,” wrote study author Jianping Hu, MD, PhD, of Fujian Medical University in Fujian, China, and colleagues.
With that in mind, the researchers evaluated the role of radiomic models based on cMRI, ADC maps, and/or SWI in predicting meningioma grade.
Patients and models
The team retrospectively analyzed 514 patients with meningioma who underwent preoperative MRI assessment over a 10-year period. There were 316 patients included in the final analysis, 229 with low-grade (grade I) and 87 with high-grade (grade II-III) meningioma.
Radiomic features from cMRI, ADC maps, and SWI were extracted based on total tumor volume.
Using a nested leave-one-out cross-validation method, the researchers evaluated the prediction performance of various radiomic models, including cMRI, ADC, SWI, ADC plus SWI, cMRI plus ADC, cMRI plus SWI, and cMRI plus ADC plus SWI.
To establish the final prediction model, the researchers used least absolute shrinkage and selection operator feature selection and implemented a random forest classifier that was trained with and without subsampling. The area under the receiver operating characteristic curve (AUC) was used to evaluate the prediction performance of each model.
Results
The model combining cMRI, ADC, and SWI had the best performance in predicting meningioma grade. The AUC of this model was 0.81 with subsampling and 0.84 without subsampling. The other models had an AUC range of 0.71-0.79 with subsampling and 0.75-0.80 without subsampling.
“Our results indicated that [the] multiparametric radiomic model based on cMRI, ADC map, and SWI [tended] to be the best model for the prediction of meningioma grade,” Dr. Hu and colleagues wrote.
Other recent studies have demonstrated that radiomic features from various imaging parameters, such as diffusion weighted imaging and cMRI, can establish robust prediction models for the prediction of meningioma grade, in which AUCs have ranged from 0.63 to 0.91.
While these findings are encouraging, the researchers acknowledged that these data should be interpreted with discretion as the cystic or necrotic areas of tumor were included in the analysis. In addition, the retrospective nature of the study could have introduced selection bias.
No funding sources were reported. The authors reported having no conflicts of interest.
SOURCE: Hu J et al. Eur J Radiol. 2020 Aug 28. doi: 10.1016/j.ejrad.2020.109251.
FROM EUROPEAN JOURNAL OF RADIOLOGY