add instructions on using contrast-agnostic weights with nnunet

quick_start_guides/use_contrast_agnostic_weights_with_nnUNet.md

@@ -0,0 +1,77 @@
+## Using contrast agnostic weights with nnUNet
+
+This guide provides instructions on how to train nnUNet models with the contrast-agnostic pre-trained weights. Ideal use cases include using the 
+pretrained weights for training/finetuning on any spinal-cord-related segmentation task (e.g. lesions, rootlets, etc.). 
+
+### Step 1: Download the pretrained weights
+
+Download the pretrained weights from the [latest release](https://github.com/sct-pipeline/contrast-agnostic-softseg-spinalcord/releases) of the 
+contrast-agnostic model. This refers to the `.zip` file of the format `model_contrast_agnostic_<date>_nnunet_compatible.zip`. 
+
+> [!WARNING]  
+> Only download the model with the `nnunet_compatible` suffix. If a release does not have this suffix, then that model weights are not directly 
+compatible with nnUNet.
+
+
+### Step 2: Create a new plans file
+
+1. Create a copy of the original `nnUNetPlans.json` file (found under `$nnUNet_preprocessed/<dataset_name_or_id>`) and 
+rename it to `nnUNetPlans_contrast_agnostic.json`. 
+2. In the `nnUNetPlans_contrast_agnostic.json`, modify the values of the following keys in the `3d_fullres` dict to be able to match the 
+values used for the contrast-agnostic model:
+
+```json
+
+"patch_size": [64, 192, 320],
+"n_stages": 6,
+"features_per_stage": [32, 64, 128, 256, 320, 320],
+"n_conv_per_stage": [2, 2, 2, 2, 2, 2],
+"n_conv_per_stage_decoder": [2, 2, 2, 2, 2],
+"strides": [
+    [1, 1, 1],
+    [2, 2, 2],
+    [2, 2, 2],
+    [2, 2, 2],
+    [2, 2, 2],
+    [1, 2, 2]
+    ],
+"kernel_sizes":[
+    [3, 3, 3],
+    [3, 3, 3],
+    [3, 3, 3],
+    [3, 3, 3],
+    [3, 3, 3],
+    [3, 3, 3]
+    ],
+"strides": [
+    [1, 1, 1],
+    [2, 2, 2],
+    [2, 2, 2],
+    [2, 2, 2],
+    [2, 2, 2],
+    [1, 2, 2]
+    ],
+
+```
+
+### Step 3: Train/Finetune the nnUNet model on your task
+
+Provide the path to the downloaded pretrained weights:
+
+```bash
+
+nnUNetv2_train <dataset_name_or_id> <configuration> <fold> -p nnUNetPlans_contrast_agnostic.json -pretrained_weights <path_to_pretrained_weights> -tr <nnUNetTrainer_Xepochs>
+
+```
+
+> [!IMPORTANT]  
+> * Training/finetuning with contrast-agnostic weights only works when for 3D nnUNet models. 
+> * Ensure that all images are in the RPI orientation before running `nnUNetv2_plan_and_preprocess`. This is because the updated 
+`patch_size` refers to patches in RPI orientation (if images are in different orientation, then the patch size might be sub-optimal).
+> * Ensure that `X` in `nnUNetTrainer_Xepochs` is set to a lower value than 1000. The idea is the finetuning does not require as 
+many epochs as training from scratch because the contrast-agnostic model has already been trained on a lot of spinal cord images 
+(so it might not require 1000 epochs to converge).
+> * The modified `nnUNetPlans_contrast_agnostic.json` might not have the same values for parameters such as `patch_size`, `strides`, 
+`n_stages`, etc. automatically set by nnUNet. As a result, the original (train-from-scratch) model might even perform better. 
+In such cases, training/finetuning with contrast-agnostic weights should just be considered as another baseline for your actual task.
+