Transfer learning with RadImageNet in medical imaging AI: review and future directions
Findings across the reviewed studies suggest that pretraining with RadImageNet produces feature representations that are more aligned with radiological image characteristics than those derived from general-purpose datasets.