Update README.md

README.md

@@ -13,8 +13,15 @@ pipeline_tag: image-segmentation
  
 ## 📑 Model Description
 
-There are all models developed using the [Children's Oncology Group AHOD 1331 dataset](https://www.nejm.org/doi/full/10.1056/NEJMoa2206660). 
-  
+There are all models developed using the [Children's Oncology Group AHOD 1331 dataset](https://www.nejm.org/doi/full/10.1056/NEJMoa2206660).
+- **CT_only**: The model only requires the planning CT images as input
+- **Early_fusion**: The model requires the planning CT, along with baseline PET/CT and interim PET/CT (post 2-cycle chemotherapy) as input; all images are deformably registered. 
+- **Late_fusion**: The model also requires the planning CT, along with baseline and interim PET/CT as inputs. The model architectures are different from those with early fusion; all images are deformable registered
+- Additional models trained in our ablation study. Note that each folder contains both early-fusion and late-fusion model weights for SwinUNETR
+  - **PET_1_2_rigid**: The SwinUNETR model with planning CT, baseline and interim PET/CT as inputs. All images are rigid registered.
+  - **PET_1_deform**: The SwinUNETR model with planning CT and baseline PET/CT (no interm PET/CT) as inputs. All images are deformably registered.
+  - **PET_1_rigid**: The SwinUNETR model with planning CT and baseline PET/CT (no interim PET/CT) as inputs. All images are deformably registered. 
+
 ## 📁 Additional Resources
 - **Codebase for model training/inference:** [GitHub](https://github.com/xtie97/ISRT-CTV-AutoSeg)
 ---