Performance comparison of 2D and 3D MRI radiomics features in meningioma grade prediction: A preliminary study

Abstract

ObjectivesThe objective of this study was to compare the predictive performance of 2D and 3D radiomics features in meningioma grade based on enhanced T1 WI images.MethodsThere were 170 high grade meningioma and 170 low grade meningioma were selected randomly. The 2D and 3D features were extracted from 2D and 3D ROI of each meningioma. The Spearman correlation analysis and least absolute shrinkage and selection operator (LASSO) regression were used to select the valuable features. The 2D and 3D predictive models were constructed by naive Bayes (NB), gradient boosting decision tree (GBDT), and support vector machine (SVM). The ROC curve was drawn and AUC was calculated. The 2D and 3D models were compared by Delong test of AUCs and decision curve analysis (DCA) curve.ResultsThere were 1143 features extracted from each ROI. Six and seven features were selected. The AUC of 2D and 3D model in NB, GBDT, and SVM was 0.773 and 0.771, 0.722 and 0.717, 0.733 and 0.743. There was no significant difference in two AUCs (p=0.960, 0.913, 0.830) between 2D and 3D model. The 2D features had a better performance than 3D features in NB models and the 3D features had a better performance than 2D features in GBDT models. The 2D features and 3D features had an equal performance in SVM models.ConclusionsThe 2D and 3D features had a comparable performance in predicting meningioma grade. Considering the issue of time and labor, 2D features could be selected for radiomics study in meningioma.Key pointsThere was a comparable performance between 2D and 3D features in meningioma grade prediction. The 2D features was a proper selection in meningioma radiomics study because of its time and labor saving.

Predictive Performance of Radiomic Features Extracted from Breast MR Imaging in Postoperative Upgrading of Ductal Carcinoma <i>in Situ</i> to Invasive Carcinoma

“…36 Although there are no studies directly comparing the performance of 2D vs 3D analysis for NME, it is supposed that observer variability is more amplified in 3D segmentation because of the need to repeatedly delineate indistinct boundaries. On the other hand, there are some reports which have compared the performance of 2D and 3D analyses in various diseases and tasks, [42][43][44][45][46][47][48][49][50] but the results have been inconsistent, and are still under debate. Among them, several studies have shown that 2D analysis had a better or equal performance than 3D analysis, suggesting that 2D segmentation can substitute 3D segmentation.…”

“…Among them, several studies have shown that 2D analysis had a better or equal performance than 3D analysis, suggesting that 2D segmentation can substitute 3D segmentation. [44][45][46][47][48][49][50] The mechanism for these results is not clear, but one possible reason is that the effect of observer variability in boundary identification is amplified in 3D, 46,47,50 rendering it more error-prone. Bos et al hypothesized that interpolation to isotropic voxels, a pre-processing step for creating a VOI, may affect the performance of 3D analysis by removing relevant feature information.…”

Voxel‐based clustered imaging by multiparameter diffusion tensor images for predicting the grade and proliferative activity of meningioma

“…Fourth, the study was limited by its small sample size of 27 patients and lack of long-term follow-up. A recent radiomics study with large sample size (Duan et al, 2023) reported the differentiation of meningioma grading, using 2D and 3D features obtained only from contrast enhanced T1-weighted images. The AUC of 2D and 3D models was 0.717-0.773, which was lower than that of our methods using DTcIs.…”

Dual-Level Augmentation Radiomics Analysis for Multisequence MRI Meningioma Grading

“…Previous literature has suggested radiomics to be promising to assist in meningioma grading, but reported performances (AUCs) widely range from 0.71 to 0.94 [11][12][13]33,[44][45][46][47][48][49][50][51][52][53]. This may be due to the majority of studies utilizing only a naïve train-test split [11][12][13]33,[44][45][46][47]49,[51][52][53] validation, with only two exiting studies reporting the average result of cross-validation (CV) folds [48,50]. However, results derived from naïve train-test splits are susceptible to selection bias and variability, and may potentially overestimate the capability of the model.…”