Table 2:

Genesis CT is pre-trained on only LUNA 2016 dataset (i.e., the source) and then fine-tuned for five distinct medical image applications (i.e., the targets). These target tasks are selected such that they show varying levels of semantic distance from the source, in terms of organs, diseases, and modalities, allowing us to investigate the transferability of the pre-trained weights of Genesis CT with respect to the domain distance. The cells checked by ✘ denote the properties that are different between the source and target datasets.

Task	Disease	Organ	Dataset	Modality
NCC
NCS
ECC	✘		✘
LCS	✘	✘	✘
BMS	✘	✘	✘	✘