Update README.md

README.md

@@ -1,59 +1,79 @@
 ---
 license: mit
 ---
+
 # Conditional Diffusion Model for Medical Image Generation
 
-This repository contains a conditional diffusion model trained to generate 3D medical CT scan images based on segmentation masks. The model uses a U-Net architecture with score-based diffusion for high-quality medical image synthesis.
+This repository contains a conditional diffusion model trained to generate **3D medical CT scan images** based on segmentation masks.  
+The model uses a **U-Net architecture with score-based diffusion** for high-quality medical image synthesis.
+
+---
 
+## Real or Fake Image?
 
-## Real or Fake Image? 
-<p>
+<p align="center">
   <img src="assets/real_fake.png" alt="Sample real vs fake medical CT" width="600"/>
 </p>
 
+---
+
+## Training Dataset
+
+The model was trained on **3,346 CT scan examples** with corresponding segmentation masks (80/20 train–validation split).  
+
+<p align="center">
+  <img src="assets/dataset.png" alt="Sample dataset" width="600"/>
+</p>
+
+**Sources:**
+1. [Kaggle Pancreas CT](https://www.kaggle.com/datasets/salihayesilyurt/pancreas-ct)  
+2. [Cancer Imaging Archive Pancreatic CT](https://nbia.cancerimagingarchive.net/nbia-search/)  
+3. [Annotated Medical Image Dataset for Segmentation Algorithms](https://drive.google.com/drive/folders/1HqEgzS8BV2c7xYNrZdEAnrHk7osJJ--2)
 
-## Trained on Dataset
-The model was trained on medical CT scan data with corresponding segmentation masks. The dataset contains 3,346 training examples with 80-20 train/validation split.
-![Sample dataset](assets/dataset.png)
-1. [Kaggle Pancreas CT](https://www.kaggle.com/datasets/salihayesilyurt/pancreas-ct)
-2. [Cancer Imaging Archive Pancreatic CT](https://nbia.cancerimagingarchive.net/nbia-search/)
-3. [A large annotated medical image dataset for the development and evaluation of segmentation algorithms](https://drive.google.com/drive/folders/1HqEgzS8BV2c7xYNrZdEAnrHk7osJJ--2)
+---
 
 ## <a href="https://archietan.com/synthetic-ct-demo" style="color:blue; text-decoration:underline;">Live Interactive Demo</a>
-<p>
-  <img src="assets/livedemo.png" alt="Sample input and output" width="600"/>
+
+<p align="center">
+  <a href="https://archietan.com/synthetic-ct-demo">
+    <img src="assets/livedemo.png" alt="Sample input and output" width="600"/>
+  </a>
 </p>
 
+---
 
 ## Model Architecture
 
-- **Base Model**: U-Net with 5-level encoder-decoder
-- **Input**: 4-channel 256x256 CT scan images
-- **Conditioning**: Segmentation masks (4-channel 256x256)
-- **Output**: 4-channel 256x256 generated images
-- **Sampling**: Euler-Maruyama sampler with 250 steps
-- **Training**: Score matching loss with conditional generation
+- **Base Model**: U-Net with 5-level encoder–decoder  
+- **Input**: 4-channel 256×256 CT scan images  
+- **Conditioning**: Segmentation masks (4-channel 256×256)  
+- **Output**: 4-channel 256×256 generated images  
+- **Sampling**: Euler–Maruyama sampler (250 steps)  
+- **Training**: Score matching loss with conditional generation  
+
+---
 
 ## Model Details
 
-- **Training Data**: 3,346 medical CT scan examples
-- **Lambda Parameter**: 25.0 (diffusion coefficient)
-- **Embedding Dimension**: 256
-- **Channels**: [32, 64, 128, 256, 512]
-- **Activation**: SiLU (Swish)
+- **Training Data**: 3,346 CT scan examples  
+- **Lambda Parameter**: 25.0 (diffusion coefficient)  
+- **Embedding Dimension**: 256  
+- **Channels**: [32, 64, 128, 256, 512]  
+- **Activation**: SiLU (Swish)  
 
-## Usage
+---
 
-This model can be used to **add more diversity to your CT-scan dataset**, which is particularly valuable when:
-- You have a **limited dataset size** (e.g., only a few hundred scans).
-- You want to **balance underrepresented anatomical variations** or rare conditions.
-- You need **synthetic augmentation** for training deep learning models in tasks such as segmentation, detection, or classification.
+## Usage
 
-### Example Applications
-- Generate additional training samples from segmentation masks to **reduce overfitting**.
-- Create synthetic CT images with controlled variations to **test model robustness**.
-- Improve representation of minority cases in the dataset to **reduce bias in medical AI**.
+This model can be used to **add more diversity to your CT-scan dataset**, especially when:
+- You have a **limited dataset size** (e.g., only a few hundred scans).  
+- You want to **balance underrepresented anatomical variations** or rare conditions.  
+- You need **synthetic augmentation** for training deep learning models in segmentation, detection, or classification.  
 
+**Example Applications**
+- Generate training samples from segmentation masks to **reduce overfitting**.  
+- Create synthetic CT images with controlled variations to **test robustness**.  
+- Improve representation of minority cases to **reduce bias in medical AI**.  
 
 ### Using the Hugging Face API
 
@@ -67,7 +87,7 @@ model = AutoModelForImageGeneration.from_pretrained("your-username/your-model-na
 # Generate images
 conditioning_mask = torch.randn(1, 4, 256, 256)  # Your segmentation mask
 generated_image = model.generate(conditioning_mask)
-```
+
 
 ### Local Usage