Derived aims and key findings of studies comparing IDH-mut and IDH wild-type LGG
| First Author and Year | Derived Aim | Key Findings |
|---|---|---|
| Fukuma 201922 | To integrate CNN deep learning features with conventional radiomic features | Conventional radiomic features: accuracy (mean ± 95% CI) = 71.7% ± 8.3%; AUC (± 95% CI) = 0.718 ± 0.139CNN features: accuracy = 69.6% ± 5.6%; AUC = 0.619 ± 0.132CNN and conventional radiomic features: accuracy = 73.1% ± 9.4%; AUC = 0.699 ± 0.145 |
| Gihr 202023 | To determine if intensity features relate to IDH status | Entropy, a second-order histogram parameter of the ADC volume was significant: IDH-mut versus IDH wild-type, mean ± SD = 5.5 ± 0.63 vs 4.75 ± 0.69; P = .0144 |
| Jakola 201824 | To determine if texture features can predict IDH status on FLAIR | Homogeneity and volume could classify IDH status with an AUC = 0.940 (85% sensitivity, 100% specificity) using the generalized linear model |
| Kim 202025 | To determine if DWI- and DSC perfusion-based image integration with standard imaging (T1WI postcontrast and FLAIR) can improve classification | Integration increased the AUC (95% CI) = 0.747 (0.663–0.832); (53.6% sensitivity and 86.7% specificity) from 0.705 (0.613–0.796) (43.9% sensitivity and 88.8% specificity) compared with conventional MR imaging radiomics |
| Li 201727 | To determine if integration of deep learning features into the radiogenomic pipeline improves classification | Conventional radiomics produced an AUC = 0.85 (sensitivity of 82.9%, specificity of 73.5%)CNN deep learning–derived features plus conventional radiomic features with feature selection produced an AUC = 0.95 (sensitivity of 94.4%; specificity of 86.7%) |
| Lu 201828 | To determine the best ML classifier | Linear SVM classified IDH status with an AUC = 0.936 (sensitivity of 85.7%, specificity of 93.0%) |
| Park 202029 | To determine if DTI improves classification when added to conventional radiomics | Addition of DTI radiomic features to conventional imaging radiomics increased the AUC (95% CI) = 0.900 (0.855–0.945) from 0.835 (0.773–0.896) |
| Ren 201930 | To compare radiomic, VASARI, and radiomic plus VASARI features derived from FLAIR, ADC, eADC, and CBF | Radiomics: AUC (95% CI) = 0.931 (0.842–1); sensitivity of 100%, specificity of 85.71%VASARI: AUC = 0.843 (sensitivity of 91.67%; specificity of 61.90%)Radiomics plus VASARI: AUC = 0.888 (0.786–0.989); sensitivity of 94.44% and specificity of 71.43% |
| Yu 201732 | To classify using the improved genetic algorithm for feature selection and leave-one-out cross-validation method in WHO grade II LGG | Using the proposed method and the SVM ML classifier, an AUC = 0.71 (sensitivity = 56% and specificity = 74%) was achieved |
| Zhou 201734 | To determine if VASARI annotations were superior to standard radiomic classification analysis | IDH classification through texture features found an AUC (± 95% CI) = 0.79 ± 0.02; sensitivity 90%, specificity of 89%IDH classification through VASARI features, AUC = 0.73 ± 0.02; sensitivity of 69%, specificity of 69% |
| Zhang 201833 | To classify by conventional radiomics | AUC = 0.830 (sensitivity = 82%, specificity = 92%) using SVM |
Note:—eADC indicates exponential ADC.