Table 2 Hyperparameters for the extraction of deep features
From: Predictive performance of radiomic models based on features extracted from pretrained deep networks
Hyperparameter | Choices |
|---|---|
Architecture | ResNet-18 |
ResNet-50 | |
VGG-19 | |
DenseNet-169 | |
Feature extraction level | Near-top |
Top | |
Segmentation | Only region of interest (ROI-full) |
Region of interest with mask as extra channel (ROI-channel) | |
Cutout region of interest (ROI-cut) | |
Slices | Slice with largest segmentation area |
All slices | |
Aggregation | Maximum |
Mean |