Update README.md

README.md

@@ -12,14 +12,16 @@ pipeline_tag: image-segmentation
 
 **Authored by**: Tie, X., Milgrom, S.A., Lo, A.C., Charpentier, A.-M., LaRiviere, M.J., Maqbool, D., Cho, S.Y., Kelly, K.M., Hodgson, D., Castellino, S.M., Hoppe, B.S., Bradshaw, T.J.
 
-[Read the full paper published in Red Journal (IJROBP)](https://www.sciencedirect.com/science/article/pii/S0360301625065927)
- 
+📄 **Related Publication**:  
+[Multi-Modality Artificial Intelligence for Involved-Site Radiation Therapy: Clinical Target Volume Delineation in High-Risk Pediatric Hodgkin Lymphoma](https://www.sciencedirect.com/science/article/pii/S0360301625065927)  
+*International Journal of Radiation Oncology · Biology · Physics (Red Journal)*
+
 ---
 ## Model Overview
 
 This repository hosts deep learning models developed for **automated clinical target volume (CTV) delineation** in **involved-site radiation therapy (ISRT)** for **high-risk pediatric Hodgkin lymphoma**.
 
-All models were trained and evaluated using imaging data from the **Children’s Oncology Group (COG) AHOD1331 phase III clinical trial**, a large, multi-institutional pediatric lymphoma dataset. The models are designed to integrate **longitudinal, multi-modality imaging** (i.e., baseline and interim PET/CT and planning CT images) to predict CTVs for radiation treatment planning. 
+All models were trained and evaluated using imaging data from the [**Children’s Oncology Group (COG) AHOD1331 phase III clinical trial**](https://www.nejm.org/doi/full/10.1056/NEJMoa2206660), a large, multi-institutional pediatric lymphoma dataset. The models are designed to integrate **longitudinal, multi-modality imaging** (i.e., baseline and interim PET/CT and planning CT images) to predict CTVs for radiation treatment planning. 
 
 ---
 
@@ -27,7 +29,7 @@ All models were trained and evaluated using imaging data from the **Children’s
 
 Depending on the model variant, inputs may include:
 
-- **Planning CT** (post-chemotherapy, treatment position)
+- **Post-Chemotherapy Planning CT**
 - **Baseline PET/CT (PET1)**  
 - **Interim PET/CT (PET2)** (after 2 cycles of chemotherapy)
 
@@ -58,6 +60,8 @@ All PET/CT images are co-registered to the planning CT using either **rigid** or
   - Fusion strategy: Late fusion using architecture-specific feature integration
   - Registration: Deformable registration for all modalities
 
+### Note that each variant has three models for ensemble.
+
 ---
 
 ### 4. Ablation Study Models (SwinUNETR)
@@ -76,14 +80,14 @@ Additional SwinUNETR models trained as part of ablation experiments are provided
   - Inputs: Planning CT + baseline PET/CT (no interim PET/CT)
   - Registration: Rigid registration
 
-Each ablation folder contains both **early-fusion** and **late-fusion** SwinUNETR model weights where applicable.
+Each ablation folder contains both **early-fusion** and **late-fusion** SwinUNETR model weights.
 
 ---
 
 ## Intended Use
 
 These models are intended for **research use only**.  
-They are designed to serve as **automated initial CTV contours** to support ISRT planning workflows and **must be reviewed and edited by qualified radiation oncologists** prior to any clinical application.
+They are designed to serve as **automated initial CTV contours** to support ISRT planning workflows and **must be reviewed and edited by radiation oncologists** prior to any clinical application.
 
 The models are **not approved for clinical decision-making** and have not undergone regulatory clearance.
 
@@ -101,11 +105,22 @@ The models are **not approved for clinical decision-making** and have not underg
 If you use these models in your research, please cite the associated publication:
 
 ```bibtex
-@article{Tie2025ISRTCTV,
-  title   = {Multi-Modality Artificial Intelligence for Involved-Site Radiation Therapy: Clinical Target Volume Delineation in High-Risk Pediatric Hodgkin Lymphoma},
-  author  = {Tie, Xin and Milgrom, Seth A. and others},
-  journal = {International Journal of Radiation Oncology Biology Physics},
-  year    = {2025}
+@article{TIE2025,
+title = {Multi-Modality Artificial Intelligence for Involved-Site Radiation Therapy: Clinical Target Volume Delineation in High-Risk Pediatric Hodgkin Lymphoma},
+journal = {International Journal of Radiation Oncology*Biology*Physics},
+year = {2025},
+issn = {0360-3016},
+doi = {https://doi.org/10.1016/j.ijrobp.2025.12.005},
+url = {https://www.sciencedirect.com/science/article/pii/S0360301625065927},
+author = {Xin Tie and Sarah A. Milgrom and Andrea C. Lo and Anne-Marie Charpentier and Michael J. LaRiviere and Danyal Maqbool and Steve Y. Cho and Kara M Kelly and David Hodgson and Sharon M. Castellino and Bradford S. Hoppe and Tyler J. Bradshaw},
+abstract = {Purpose
+Clinical target volume (CTV) delineation for involved-site radiation therapy (ISRT) in Hodgkin lymphoma (HL) is time-consuming due to the need to analyze multi-time-point PET/CT scans co-registered to the planning CT. Our goal was to develop automated CTV segmentation algorithms that integrated multi-modality imaging to facilitate ISRT planning.
+Methods
+This study included planning CT, baseline PET/CT (PET1), and interim PET/CT (PET2) scans from 288 pediatric patients with high-risk HL enrolled in the [redacted] trial. Data from 58 patients across 24 institutions were held out for external testing, while the remaining 230 cases from 95 institutions were used for model development. We investigated three deep learning (DL) architectures (SegResNet, ResUNet, and SwinUNETR) and evaluated the impact of incorporating PET1 and PET2 images along with the planning CT. Performance was assessed using the Dice similarity coefficient (DSC) and 95% Hausdorff distance (HD95). Inter-observer variability (IOV) was estimated by comparing original institutional CTVs with those newly delineated by four radiation oncologists on 10 cases. The quality of CTVs generated by the top-performing model was independently assessed by radiation oncologists on 40 other cases using a 5-point Likert scale and compared against the original institutional CTVs.
+Results
+On the external cohort, a SwinUNETR model incorporating planning CT, PET1, and PET2 images achieved the highest performance, with a DSC of 0.72 and HD95 of 34.43 mm. All models incorporating PET/CT images were significantly better (P<0.01) than planning CT-only models. IOV analysis yielded a DSC of 0.70 and HD95 of 30.14 mm. In clinical evaluation, DL-generated CTVs received a mean quality score of 3.38 out of 5, comparable to original physician-delineated CTVs (3.13; P = 0.13)
+Conclusion
+The DL model was able to generate clinically useful CTVs with quality comparable to manually delineated CTVs, suggesting its potential to improve physician efficiency in ISRT planning.}
 }