UNETR implementation incompatible with MONAI >= 1.3.0 due to deprecated pos_embed parameter

[perctrix]

Describe the bug The UNETR implementation in the research-contributions repository uses an outdated parameter name pos_embed that is incompatible with MONAI versions >= 1.3.0. Starting from MONAI 1.3.0, the parameter was renamed from pos_embed to proj_type in the ViT class.

To Reproduce Steps to reproduce the behavior:

1. Install MONAI >= 1.3.0 (tested with MONAI 1.5.0)
2. Clone the research-contributions repository
3. Navigate to UNETR/BTCV/networks/unetr.py
4. Try to instantiate the UNETR model:

from networks.unetr import UNETR
model = UNETR(in_channels=1, out_channels=14, img_size=(96, 96, 96))

5. Observe the error when ViT is initialized with pos_embed parameter

Expected behavior The UNETR model should work with current MONAI versions without parameter compatibility issues. The parameter should be proj_type for MONAI >= 1.3.0.

Screenshots N/A

Environment (please complete the following information):

• OS: Windows 11
• Python version: 3.12.8
• MONAI version: 1.5.0
• PyTorch version: 2.8.0+cu129
• CUDA/cuDNN version: CUDA 12.9
• GPU models and configuration: NVIDIA GeForce RTX 4060 Laptop GPU

Additional context The issue affects users trying to use the UNETR implementation with newer MONAI versions. A simple fix would be to:

1. Update line 37 from pos_embed: str = "perceptron" to proj_type: str = "perceptron"
2. Update line 96 from pos_embed=pos_embed to proj_type=proj_type
3. Update the docstring and examples accordingly

This change would maintain backward compatibility if the minimum supported MONAI version is updated to 1.3.0, or alternatively, a version check could be implemented to support both parameter names.