Predicting occult lymph node metastasis in solid-predominantly invasive lung adenocarcinoma across multiple centers using radiomics-deep learning fusion model

Table 3 Performance comparison of different feature fusion techniques on the internal validation set, independent test set 1, and independent test set 2

Model	Feature Fusion	Internal Validation Set	Independent Test Set 1	Independent Test Set 2	aAUC (SD)
Radiomics	None	0.810	0.644	0.679	0.715 (0.019)
Deep Learning	None	0.746	0.717	0.623	0.676 (0.021)
Radiomics-Deep Learning Fusion	Addition^a	0.795	0.663	0.736	0.736 (0.018)
	Learnable Addition^a	0.793	0.677	0.748	0.740 (0.017)
	Concatenation^a	0.790	0.697	0.743	0.737 (0.018)
	Concatenation	0.813	0.758	0.728	0.754 (0.018)

Performance was reported in terms of AUC. Radiomics employed the support vector machine model, while deep learning utilized the pre-trained ResNet-18 model
^aBefore feature fusion, the radiomics features were mapped to the same dimension as the deep learning features

ISSN: 1470-7330