Fig. 2

Design of the study. A Overall framework of the radiomics-deep learning fusion model for the prediction of OLNM in SPILAC. The w/o stands for with/without. \({\mathcal{L}}_{ce}\) and \({\mathcal{L}}_{cl}\) represent the cross-entropy loss function and the contrastive learning loss function, respectively. B Comparison of different feature fusion techniques. ^* denotes the mapping of radiomics features to the same dimension as the deep learning features through the fully connected layer. \({w}_{1}\) and \({w}_{2}\) represent the learnable weight parameters