update readme

Author: mrokuss

readme.md

@@ -1,144 +1,107 @@
-# Welcome to the new nnU-Net!
-
-Click [here](https://github.com/MIC-DKFZ/nnUNet/tree/nnunetv1) if you were looking for the old one instead.
-
-Coming from V1? Check out the [TLDR Migration Guide](documentation/tldr_migration_guide_from_v1.md). Reading the rest of the documentation is still strongly recommended ;-)
-
-## **2024-04-18 UPDATE: New residual encoder UNet presets available!**
-Residual encoder UNet presets substantially improve segmentation performance.
-They ship for a variety of GPU memory targets. It's all awesome stuff, promised! 
-Read more :point_right: [here](documentation/resenc_presets.md) :point_left:
-
-Also check out our [new paper](https://arxiv.org/pdf/2404.09556.pdf) on systematically benchmarking recent developments in medical image segmentation. You might be surprised!
-
-# What is nnU-Net?
-Image datasets are enormously diverse: image dimensionality (2D, 3D), modalities/input channels (RGB image, CT, MRI, microscopy, ...), 
-image sizes, voxel sizes, class ratio, target structure properties and more change substantially between datasets. 
-Traditionally, given a new problem, a tailored solution needs to be manually designed and optimized  - a process that 
-is prone to errors, not scalable and where success is overwhelmingly determined by the skill of the experimenter. Even 
-for experts, this process is anything but simple: there are not only many design choices and data properties that need to 
-be considered, but they are also tightly interconnected, rendering reliable manual pipeline optimization all but impossible! 
-
-![nnU-Net overview](documentation/assets/nnU-Net_overview.png)
-
-**nnU-Net is a semantic segmentation method that automatically adapts to a given dataset. It will analyze the provided 
-training cases and automatically configure a matching U-Net-based segmentation pipeline. No expertise required on your 
-end! You can simply train the models and use them for your application**.
-
-Upon release, nnU-Net was evaluated on 23 datasets belonging to competitions from the biomedical domain. Despite competing 
-with handcrafted solutions for each respective dataset, nnU-Net's fully automated pipeline scored several first places on 
-open leaderboards! Since then nnU-Net has stood the test of time: it continues to be used as a baseline and method 
-development framework ([9 out of 10 challenge winners at MICCAI 2020](https://arxiv.org/abs/2101.00232) and 5 out of 7 
-in MICCAI 2021 built their methods on top of nnU-Net, 
- [we won AMOS2022 with nnU-Net](https://amos22.grand-challenge.org/final-ranking/))!
-
-Please cite the [following paper](https://www.google.com/url?q=https://www.nature.com/articles/s41592-020-01008-z&sa=D&source=docs&ust=1677235958581755&usg=AOvVaw3dWL0SrITLhCJUBiNIHCQO) when using nnU-Net:
-
-    Isensee, F., Jaeger, P. F., Kohl, S. A., Petersen, J., & Maier-Hein, K. H. (2021). nnU-Net: a self-configuring 
-    method for deep learning-based biomedical image segmentation. Nature methods, 18(2), 203-211.
-
-
-## What can nnU-Net do for you?
-If you are a **domain scientist** (biologist, radiologist, ...) looking to analyze your own images, nnU-Net provides 
-an out-of-the-box solution that is all but guaranteed to provide excellent results on your individual dataset. Simply 
-convert your dataset into the nnU-Net format and enjoy the power of AI - no expertise required!
-
-If you are an **AI researcher** developing segmentation methods, nnU-Net:
-- offers a fantastic out-of-the-box applicable baseline algorithm to compete against
-- can act as a method development framework to test your contribution on a large number of datasets without having to 
-tune individual pipelines (for example evaluating a new loss function)
-- provides a strong starting point for further dataset-specific optimizations. This is particularly used when competing 
-in segmentation challenges
-- provides a new perspective on the design of segmentation methods: maybe you can find better connections between 
-dataset properties and best-fitting segmentation pipelines?
-
-## What is the scope of nnU-Net?
-nnU-Net is built for semantic segmentation. It can handle 2D and 3D images with arbitrary 
-input modalities/channels. It can understand voxel spacings, anisotropies and is robust even when classes are highly
-imbalanced.
-
-nnU-Net relies on supervised learning, which means that you need to provide training cases for your application. The number of 
-required training cases varies heavily depending on the complexity of the segmentation problem. No 
-one-fits-all number can be provided here! nnU-Net does not require more training cases than other solutions - maybe 
-even less due to our extensive use of data augmentation. 
-
-nnU-Net expects to be able to process entire images at once during preprocessing and postprocessing, so it cannot 
-handle enormous images. As a reference: we tested images from 40x40x40 pixels all the way up to 1500x1500x1500 in 3D 
-and 40x40 up to ~30000x30000 in 2D! If your RAM allows it, larger is always possible.
-
-## How does nnU-Net work?
-Given a new dataset, nnU-Net will systematically analyze the provided training cases and create a 'dataset fingerprint'. 
-nnU-Net then creates several U-Net configurations for each dataset: 
-- `2d`: a 2D U-Net (for 2D and 3D datasets)
-- `3d_fullres`: a 3D U-Net that operates on a high image resolution (for 3D datasets only)
-- `3d_lowres` → `3d_cascade_fullres`: a 3D U-Net cascade where first a 3D U-Net operates on low resolution images and 
-then a second high-resolution 3D U-Net refined the predictions of the former (for 3D datasets with large image sizes only)
-
-**Note that not all U-Net configurations are created for all datasets. In datasets with small image sizes, the 
-U-Net cascade (and with it the 3d_lowres configuration) is omitted because the patch size of the full 
-resolution U-Net already covers a large part of the input images.**
-
-nnU-Net configures its segmentation pipelines based on a three-step recipe:
-- **Fixed parameters** are not adapted. During development of nnU-Net we identified a robust configuration (that is, certain architecture and training properties) that can 
-simply be used all the time. This includes, for example, nnU-Net's loss function, (most of the) data augmentation strategy and learning rate.
-- **Rule-based parameters** use the dataset fingerprint to adapt certain segmentation pipeline properties by following 
-hard-coded heuristic rules. For example, the network topology (pooling behavior and depth of the network architecture) 
-are adapted to the patch size; the patch size, network topology and batch size are optimized jointly given some GPU 
-memory constraint. 
-- **Empirical parameters** are essentially trial-and-error. For example the selection of the best U-net configuration 
-for the given dataset (2D, 3D full resolution, 3D low resolution, 3D cascade) and the optimization of the postprocessing strategy.
-
-## How to get started?
-Read these:
-- [Installation instructions](documentation/installation_instructions.md)
-- [Dataset conversion](documentation/dataset_format.md)
-- [Usage instructions](documentation/how_to_use_nnunet.md)
-
-Additional information:
-- [Learning from sparse annotations (scribbles, slices)](documentation/ignore_label.md)
-- [Region-based training](documentation/region_based_training.md)
-- [Manual data splits](documentation/manual_data_splits.md)
-- [Pretraining and finetuning](documentation/pretraining_and_finetuning.md)
-- [Intensity Normalization in nnU-Net](documentation/explanation_normalization.md)
-- [Manually editing nnU-Net configurations](documentation/explanation_plans_files.md)
-- [Extending nnU-Net](documentation/extending_nnunet.md)
-- [What is different in V2?](documentation/changelog.md)
-
-Competitions:
-- [AutoPET II](documentation/competitions/AutoPETII.md)
-
-[//]: # (- [Ignore label](documentation/ignore_label.md))
-
-## Where does nnU-Net perform well and where does it not perform?
-nnU-Net excels in segmentation problems that need to be solved by training from scratch, 
-for example: research applications that feature non-standard image modalities and input channels,
-challenge datasets from the biomedical domain, majority of 3D segmentation problems, etc . We have yet to find a 
-dataset for which nnU-Net's working principle fails!
-
-Note: On standard segmentation 
-problems, such as 2D RGB images in ADE20k and Cityscapes, fine-tuning a foundation model (that was pretrained on a large corpus of 
-similar images, e.g. Imagenet 22k, JFT-300M) will provide better performance than nnU-Net! That is simply because these 
-models allow much better initialization. Foundation models are not supported by nnU-Net as 
-they 1) are not useful for segmentation problems that deviate from the standard setting (see above mentioned 
-datasets), 2) would typically only support 2D architectures and 3) conflict with our core design principle of carefully adapting 
-the network topology for each dataset (if the topology is changed one can no longer transfer pretrained weights!) 
-
-## What happened to the old nnU-Net?
-The core of the old nnU-Net was hacked together in a short time period while participating in the Medical Segmentation 
-Decathlon challenge in 2018. Consequently, code structure and quality were not the best. Many features 
-were added later on and didn't quite fit into the nnU-Net design principles. Overall quite messy, really. And annoying to work with.
-
-nnU-Net V2 is a complete overhaul. The "delete everything and start again" kind. So everything is better 
-(in the author's opinion haha). While the segmentation performance [remains the same](https://docs.google.com/spreadsheets/d/13gqjIKEMPFPyMMMwA1EML57IyoBjfC3-QCTn4zRN_Mg/edit?usp=sharing), a lot of cool stuff has been added. 
-It is now also much easier to use it as a development framework and to manually fine-tune its configuration to new 
-datasets. A big driver for the reimplementation was also the emergence of [Helmholtz Imaging](http://helmholtz-imaging.de), 
-prompting us to extend nnU-Net to more image formats and domains. Take a look [here](documentation/changelog.md) for some highlights.
-
-# Acknowledgements
-<img src="documentation/assets/HI_Logo.png" height="100px" />
-
-<img src="documentation/assets/dkfz_logo.png" height="100px" />
-
-nnU-Net is developed and maintained by the Applied Computer Vision Lab (ACVL) of [Helmholtz Imaging](http://helmholtz-imaging.de) 
-and the [Division of Medical Image Computing](https://www.dkfz.de/en/mic/index.php) at the 
-[German Cancer Research Center (DKFZ)](https://www.dkfz.de/en/index.html).
+# [MICCAI2025 Panther] Multi-Stage Fine-Tuning and Ensembling for Pancreatic Tumor Segmentation in MRI
+
+This repository contains the official implementation of our MICCAI 2025 challenge paper:
+
+### **A Multi-Stage Fine-Tuning and Ensembling Strategy for Pancreatic Tumor Segmentation in Diagnostic and Therapeutic MRI**
+
+We present a **cascaded pre-training, fine-tuning, and ensembling framework** for **pancreatic ductal adenocarcinoma (PDAC) segmentation** in MRI. Our method, built on **nnU-Net v2**, leverages:  
+- Multi-stage cascaded fine-tuning (from a foundation model → CT lesions → target MRI modalities)  
+- Systematic augmentation ablations (aggressive vs. default)  
+- Metric-aware heterogeneous ensembling ("mix of experts")  
+
+This approach achieved **the first place** in the **PANTHER challenge**, with robust performance across both **diagnostic T1W (Task 1)** and **therapeutic T2W MR-Linac (Task 2)** scans.
+
+> **Authors**: Omer Faruk Durugol*, Maximilian Rokuss*, Yannick Kirchhoff, Klaus H. Maier-Hein  
+> *Equal contribution  
+> **Paper**: [![arXiv](https://img.shields.io/badge/arXiv-2508.21775-b31b1b.svg)](https://arxiv.org/abs/2508.21775)  
+> **Challenge**: [PANTHER](https://panther.grand-challenge.org)
+
+---
+
+## News/Updates
+- 🏆 **Aug 2025**: Achieved **1st place in both tasks** in the [PANTHER Challenge](https://panther.grand-challenge.org)!  
+- 📄 **Aug 2025**: Paper preprint released on [arXiv](https://arxiv.org/abs/2508.21775).  
+
+---
+
+## 🚀 Usage
+
+This section provides a complete guide to reproduce our training and inference workflow. Our method is developed entirely within the nnU-Net v2 framework.
+
+### Installation and Setup
+   
+First, set up an environment with PyTorch and clone this repository to access our custom code. Here, we refer to nnU-Net v2 repository [installation instructions](https://github.com/MIC-DKFZ/nnUNet/blob/master/documentation/installation_instructions.md) for a detailed explanation but provide a rundown on how to set up the environment quickly:
+
+#### 1. Create and activate your Conda environment (recommended)
+```python
+conda create -n panther python=3.9
+conda activate panther
+```
+
+#### 2. Install [PyTorch](https://pytorch.org/get-started/locally) as described on their website (conda/pip)
+
+#### 3. Clone this repository
+```bash
+git clone https://github.com/MIC-DKFZ/panther.git
+cd panther
+```
+
+#### 4. Install nnU-Net v2 and our project in editable mode
+This makes our custom trainers and plans available to the framework.
+```python
+pip install -e .
+```
+
+#### 5. Set environment variables
+We recommend editing the `.bashrc` (or corresponding startup shell) file of your home folder by adding the following lines:
+```python
+export nnUNet_raw="/panther/nnUNet_raw"
+export nnUNet_preprocessed="/panther/nnUNet_preprocessed"
+export nnUNet_results="/panther/nnUNet_results"
+```
+(Change the paths according to your preferred directory structure.)
+
+### Pre-trained Checkpoints
+
+Coming soon!
+
+### Evaluation
+
+Coming soon!
+
+---
+
+## 📂 Data
+
+We trained and evaluated on the **PANTHER Challenge dataset**:
+👉 [https://zenodo.org/records/15192302](https://zenodo.org/records/15192302)
+
+Pretraining leveraged:
+
+* [MultiTalentV2](https://zenodo.org/records/13753413)
+* Pancreatic lesion CT datasets (MSD + PANORAMA)
+
+---
+
+## 📚 Citation
+
+If you find this repository useful, please cite:
+
+```bibtex
+@article{durugol2025multistagefinetuningensemblingstrategy,
+      title={A Multi-Stage Fine-Tuning and Ensembling Strategy for Pancreatic Tumor Segmentation in Diagnostic and Therapeutic MRI}, 
+      author={Omer Faruk Durugol and Maximilian Rokuss and Yannick Kirchhoff and Klaus H. Maier-Hein},
+      year={2025},
+      eprint={2508.21775},
+      archivePrefix={arXiv},
+      primaryClass={cs.CV},
+      url={https://arxiv.org/abs/2508.21775}, 
+}
+```
+
+---
+
+## 📬 Contact
+
+For questions, issues, or collaborations, please reach out:
+📧 [maximilian.rokuss@dkfz-heidelberg.de](mailto:maximilian.rokuss@dkfz-heidelberg.de)