ASLDM

Code for "Anatomy-aware Sketch-guided Latent Diffusion Model for Orbital Tumor Multi-Parametric MRI Missing Modalities Synthesis"

This repository contains code and example data for training and testing the ASLDM model on the OTTS dataset, enabling synthesis of missing modalities from multi-parametric MRI using anatomical sketch guidance.

---

🧠 OTTS Example Data & Pre-trained Weights

Please download the two pre-trained model weight files from the following Google Drive links:

• ASLDM Autoencoder Weights
• ASLDM UNet Weights

After downloading, place these weight files into the checkpoints/ directory of the project. For example:

ASLDM/
├── checkpoints/
│   ├── autoencoder.pt
│   └── diffusion_unet.pt

⚠️ Note: Ensure the weight file names match those specified in the code to avoid loading errors.

---

Below are examples of each modality used in our model:

T1WI Anatomical structure	T2WI Edema/fluid sensitivity	T1CE Contrast-enhanced lesion	DWI Diffusion signal
ADC Quantitative diffusion	Seg Tumor mask	Sketch Structural prior

---

⚙️ Setup Instructions

1. Environment Setup

We recommend using Python 3.9+ and creating a virtual environment:

conda create -n asldm python=3.12
conda activate asldm

2. Install Dependencies

pip install -r requirements.txt

Additional dependencies might include:

• torch >= 2.0
• torchvision
• monai
• scikit-image
• pillow
• numpy
• tqdm
• itertools
• tqdm
• natsort

3. Dataset

Place the data in the data/test/ directory. File names should follow:

<data_root>/<modality>/<sample_id>-<modality>-slice_<slice_id>.png

---

🚀 Inference Example

Run inference with a sketch input and available modalities:

python test.py 

---

📂 Folder Structure

ASLDM/
├── data/
│   └── test/
│       ├── t1n/
│       ├── t2w/
│       ├── t1c/
│       ├── dwi/
│       ├── adc/
│       ├── seg/
│       └── sketch/
├── models/
├── test.py
├── train.py
└── README.md