README.md

---
title: bfd-rg
sdk: gradio
emoji: 🏃
colorFrom: green
colorTo: yellow
pinned: false
---
![Add a heading.png](https://cdn-uploads.huggingface.co/production/uploads/65bb837dbfb878f46c77de4c/AubiFwkdFgFgN6KfHIVhb.png)

# **Bone-Fracture-Detection**

> **Bone-Fracture-Detection** is a binary image classification model based on `google/siglip2-base-patch16-224`, trained to detect **fractures in bone X-ray images**. It is designed for use in **medical diagnostics**, **clinical triage**, and **radiology assistance systems**.


```py
Classification Report:
               precision    recall  f1-score   support

    Fractured     0.8633    0.7893    0.8246      4480
Not Fractured     0.8020    0.8722    0.8356      4383

     accuracy                         0.8303      8863
    macro avg     0.8326    0.8308    0.8301      8863
 weighted avg     0.8330    0.8303    0.8301      8863
 ```

![download.png](https://cdn-uploads.huggingface.co/production/uploads/65bb837dbfb878f46c77de4c/aoLW8h2vfmEPH60676rnb.png)


---

## **Label Classes**

The model distinguishes between the following bone conditions:

```
0: Fractured  
1: Not Fractured
```

---

## **Installation**

```bash
pip install transformers torch pillow gradio
```

---

## **Example Inference Code**

```python
import gradio as gr
from transformers import AutoImageProcessor, AutoModelForImageClassification
from PIL import Image
import torch

# Load model and processor from the Hugging Face Hub
model_name = "prithivMLmods/Bone-Fracture-Detection"
model = AutoModelForImageClassification.from_pretrained(model_name)
processor = AutoImageProcessor.from_pretrained(model_name)

def detect_fracture(image):
    """
    Takes a NumPy image array, processes it, and returns the model's prediction.
    """
    # Convert NumPy array to a PIL Image
    image = Image.fromarray(image).convert("RGB")
    
    # Process the image and prepare it as input for the model
    inputs = processor(images=image, return_tensors="pt")

    # Perform inference without calculating gradients
    with torch.no_grad():
        outputs = model(**inputs)
        logits = outputs.logits
        
        # Apply softmax to get probabilities and convert to a list
        probs = torch.nn.functional.softmax(logits, dim=1).squeeze().tolist()

    # Create a dictionary of labels and their corresponding probabilities
    # This now correctly uses the labels from the model's configuration
    prediction = {model.config.id2label[i]: round(probs[i], 3) for i in range(len(probs))}
    
    return prediction

# Create the Gradio Interface
iface = gr.Interface(
    fn=detect_fracture,
    inputs=gr.Image(type="numpy", label="Upload Bone X-ray"),
    outputs=gr.Label(num_top_classes=2, label="Detection Result"),
    title="🔬 Bone Fracture Detection",
    description="Upload a bone X-ray image to detect if there is a fracture. The model will return the probability for 'Fractured' and 'Not Fractured'.",
    examples=[
        ["fractured_example.png"],
        ["not_fractured_example.png"]
    ] # Note: You would need to have these image files in the same directory for the examples to work.
)

# Launch the app
if __name__ == "__main__":
    iface.launch()

```

---

## **Applications**

* **Orthopedic Diagnostic Support**
* **Emergency Room Triage**
* **Automated Radiology Review**
* **Clinical Research in Bone Health**