Table 2 Overview of the predictive performance reported in the included studies
From: Are deep models in radiomics performing better than generic models? A systematic review
- If only a cross-validation was performed without an independent internal validation cohort, the number of training samples across all folds is reported the #CV column. Otherwise, if internal validation cohort was available, then the number of training and validation samples are reported in the #training and #test column. Note that Li (2020), Astaraki (2021), Caballo (2021), Chen + Lin (2021), Li (2021), Lin (2021), and Xuan (2021) used data from external sites; however, the data were merged and randomly split before modeling. Therefore, the results were considered to be internally validated, not externally. Note also that Hu + Gong (2021) and Song + Wang + Luo (2021) use a U-Net, which is a generative network.
yes,
no, 2D Two-dimensional, 3D Three-dimensional, CapsNet Capsule neural network, CNN Convolutional neural network, GAN Generative adversarial network, SAE Sparse autoencoder
yes,
no, 2D Two-dimensional, 3D Three-dimensional, CapsNet Capsule neural network, CNN Convolutional neural network, GAN Generative adversarial network, SAE Sparse autoencoder