Table 5 Test performances of all methods on multiple domains are presented. The mean and standard deviation across five random seeds are shown. The overall performance is the weighted average AUROC by the number of data in each site. Bold is the best and underlined is the second best

Supervised (baseline)

 Single data

0.994 ± 0.001

0.930 ± 0.003

0.874 ± 0.008

0.867 ± 0.019

 Merged data

0.977 ± 0.003

0.842 ± 0.023

0.916 ± 0.002

0.924 ± 0.007

0.929 ± 0.005

Federated

 FedAvg

0.980 ± 0.004

0.751 ± 0.013

0.901 ± 0.007

0.876 ± 0.013

0.901 ± 0.003

 FedProx

0.984 ± 0.001

0.735 ± 0.010

0.906 ± 0.004

0.882 ± 0.007

0.900 ± 0.003

Finetuning

 Small to Large

0.994 ± 0.000

0.584 ± 0.017

0.829 ± 0.013

0.768 ± 0.017

0.845 ± 0.007

 Large to Small

0.839 ± 0.055

0.856 ± 0.024

0.874 ± 0.023

0.939 ± 0.003

0.856 ± 0.027

Continual

 Small to Large

0.935 ± 0.010

0.785 ± 0.023

0.876 ± 0.012

0.871 ± 0.013

0.883 ± 0.006

 Large to Small

0.908 ± 0.020

0.873 ± 0.026

0.896 ± 0.004

0.912 ± 0.007

0.897 ± 0.005