Image Segmentation & Detection Web Application

This Streamlit-powered application offers an interactive platform for analyzing medical images using detection and segmentation tasks.

Can upload medical images (e.g., MRI, CT, TRUS), select a model architecture (currently placeholder), perform tasks, and visualize results. While this version contains dummy logic for modeling. It is structured for integration with actual deep learning models.

---

In Progress.,.,.

---

Features

• Upload MRI, CT, or TRUS images in .png, .jpeg, .tiff, or .nifti format
• Select model type (GAN, YOLOv8, Diffusion - placeholders)
• Choose task: Detection, Segmentation, or Both
• View overlays for segmentation and bounding boxes
• Upload ground truth masks to compute evaluation metrics (IoU, Dice)
• Download predicted segmentation masks
• Fully interactive web-based experience via Streamlit

---

Project Structure

.
├── streamlit_app.py              # Main Streamlit UI script
├── saved_predictions/           # Folder to save segmentation outputs
└── README.md

---

Installation

Install required dependencies using pip:

pip install streamlit torch torchvision pillow opencv-python-headless scikit-learn

---

How to Run

Launch the Streamlit interface using the following command:

streamlit run streamlit_app.py

---

Workflow

Step 1: Upload Files

• Upload a medical image from your local machine
• Optionally upload a ground truth mask to compute metrics

Step 2: Select Parameters

• Choose a model type (placeholder options provided)
• Choose your task (Detection / Segmentation / Both)
• Select the image modality (MRI / CT / TRUS)

Step 3: Run Task

• Click " Run Task" to start the process
• Outputs are rendered side-by-side (original, prediction)
• Download the predicted mask (PNG format)
• View metrics if a ground truth mask is uploaded

---

Evaluation Metrics

If a ground truth mask is uploaded, the app calculates:

Metric	Description
IoU (Jaccard)	Measures overlap between predicted and ground truth mask
Dice Score	Measures similarity between predicted and ground truth regions

---

Output Example

• Bounding box visualization for detection
• Overlayed mask output for segmentation
• Score summary in metric section
• Downloadable mask result

---

Extending the App

To integrate your trained models, update the following placeholder functions:

def load_model(model_name): ...
def generate_mask(image, model): ...
def detect_regions(image, model): ...

These can be customized to include TensorFlow, PyTorch, or ONNX inference pipelines...

---

Limitations

• Models used are placeholders (i.e., no learning or inference)
• Segmentation and detection results are simulated for demo purposes
• NIfTI file format requires additional packages like nibabel

---

Author

Developed as a modular prototype for interactive medical image processing applications.

---

License

This project is intended for research and educational use. For clinical deployment, consult relevant medical regulations and validation standards. This project is licensed under the GNU General Public License v3.0 - see the LICENSE file for details.