Inquiry About Input Channel Preprocessing for RadImageNet Pretrained Models

[Fisherman27]

Thank you very much for open-sourcing RadImageNet — it has been incredibly helpful for my current research!
While using the PyTorch pretrained model (ResNet50), I noticed that the input channels are set to 3. However, medical images such as CT scans are typically single-channel. To better utilize RadImageNet, I would like to ask: during pretraining, how did you convert single-channel images into 3-channel inputs? Was it done by repeating the single channel (e.g., stacking)?
Your clarification would be extremely helpful for my work. Thank you again for your excellent contribution and support!