Improving accuracy of 3d image classification model

I made an image classification model for image feature extraction. My model is a 3d model. Because I couldn’t find any 3d pretrained model. I trained the model from scratch. First I didn’t get good results then I tried to use data augmentation and now I got results that are around 65 percent but I needed higher accuracy. Do you have any idea that how can I improve accuracy results in 3d image classification?
The inputs to my model are Kidney MRI. I tried to find a dataset that has a lot of 3d medical images and use transfer learning but I couldn’t find a huge dataset like imagenet. Is there any way to use image net in 3d models. Or is there any way to load weights of 2d models to 3d or vice versa.
If anyone has experience in this regard, please help and share your ideas.