Table 3.
Mean training and validation scores for the best performing machine learning models using the 4.0 SUV threshold segmentation technique.
| Machine Learning Model | Hyperparameters | Features | AUC Mean Training | AUC Mean Validation |
|---|---|---|---|---|
| SUVmax/130 | ||||
| Ridge Regression | C: 1 × 10−5, penalty: l2, solver: liblinear | Stage One, PET wavelet-LLH GLSZM Large Area Emphasis, PET wavelet-HHH GLSZM Grey Level Non-Uniformity Normalised, PET square 10th Percentile, PET square GLDM Grey Level Non-Uniformity | 0.75 (0.02) | 0.74 (0.07) |
| Support Vector Machine | C: 1, gamma: 0.008915428868611115, kernel: sigmoid | PET wavelet-HHH GLSZM Grey Level Non-Uniformity Normalised, PET square 10th Percentile, PET lbp-3D-m1 Interquartile Range, PET lbp-3D-m1 GLDM Large Dependence Low Grey Level Emphasis, PET lbp-3D-k 90th Percentile | 0.74 (0.02) | 0.73 (0.07) |
| Random Forest | bootstrap: False, max depth: 1, max features: log2, min samples leaf: 50, min samples split: 50, n estimators: 10 | PET original shape Maximum 2D Diameter Column, MTV, PET original first order Kurtosis, PET original GLSZM Large Area Emphasis, PET wavelet-LHL GLCM Correlation, PET wavelet-LHL GLCM Imc2 | 0.76 (0.02) | 0.71 (0.08) |
| SUVmax/64 | ||||
| Ridge Regression | C: 0.001, penalty: l2, solver: newton-cg | Stage Four, PET original GLSZM Large Area Emphasis, PET wavelet-HHL GLSZM Small Area Emphasis, PET wavelet-HHH GLSZM Grey Level Non-Uniformity Normalised, PET square 10th Percentile | 0.77 (0.02) | 0.75 (0.06) |
| Support Vector Machine | C: 0.1, gamma: 0.07938667031015477, kernel: rbf | PET original GLDM Large Dependence Low Grey Level Emphasis, PET wavelet-HHH GLSZM Grey Level Non-Uniformity Normalised, PET square 10th Percentile, PET lbp-3D-k 90 Percentile, PET lbp-3D-k GLSZM Size Zone Non-Uniformity Normalised | 0.75 (0.02) | 0.72 (0.06) |
| Random Forest | bootstrap: True, max depth: 1, max features: log2, min samples leaf: 44, min samples split: 6, n estimators: 243 | PET original shape Maximum 2D Diameter Column, PET original shape Surface Volume Ratio, PET original 10th Percentile | 0.71 (0.02) | 0.69 (0.08) |
l2 = Ridge regression penalty, liblinear = A library for large linear classification, GLSZM = grey level size zone matrix, GLDM = grey level dependence matrix, lbp-3D-m1 = local binary pattern filtered image at level 1, lbp-3D-k = local binary pattern kurtosis image, GLCM = grey level co-occurrence matrix, rbf = radial basis function.